Empowering Breakthrough Innovations Through Digital Technology: The Effects of Digital Technology Depth and Breadth

Abstract

The advent of digital technology has empowered firms to make breakthrough innovations; however, how digital technology adoption (especially its depth and breadth) influences such innovations remains unknown. This study addresses this gap by analyzing panel data from 469 listed biomedical companies in China from 2010 to 2021 using a knowledge-based view (KBV). Results show the following: (1) both dimensions of digital technology adoption are positively correlated with firms’ breakthrough innovations; (2) firms’ knowledge recombination novelty mediates these relationships; (3) the mediating effect of knowledge recombination novelty diminishes as firms’ knowledge base increases. Additional analysis indicates that the positive impacts of digital technology depth and breadth on breakthrough innovations are stronger for enterprises located in cities with high digital infrastructure construction levels (high-DICs) than enterprises located in low (and medium)-DIC cities. These findings enhance existing research on the mechanisms and boundary conditions of digital technology-driven breakthrough innovation and offer new insights for enterprises to gain sustainable competitive advantage in the digital era.

1. Introduction

Breakthrough innovations are typically viewed as ruptures along specific technological trajectories, potentially resulting in shifts or transformations within the dominant technological paradigm [1,2]. As a crucial component of the process of “creative destruction”, breakthrough innovations play an essential role in reshaping the competitive landscape and occupying a profitable market position [3,4]. For firms, developing successful breakthrough innovations is a challenge due to the high degree of uncertainty and complexity involved [5,6]. This difficulty is further heightened by the rapid rise and development of emerging technologies, which has led to a surge in the frequency and intensity of industry technology evolution, compelling firms to adopt more radical approaches to innovation [7]. As an organizational resource, digital technology has the characteristics of data homogeneity, reprogrammability, and affordance [8]. By integrating digital technologies with business and management processes, firms can effectively broaden information channels, improve the efficiency of technological knowledge flow, and, thus, increase the likelihood of achieving breakthrough innovations [9]. Therefore, exploring how enterprises leverage digital technologies to improve breakthrough innovation performance is of great practical significance in the digital era.

To date, although research on digital technology and firm innovation is still flourishing [10,11], there remains a lack of consensus regarding the impact between the two [12]. For example, Radicic and Petkovic [13] found that digitalization enhances enterprises’ business models, competitive advantages, and innovation performance. In contrast, Usai et al. [14] reported that digital technologies exert a minimal impact on firms’ innovation performance, with R&D expenditures serving as the most reliable predictor of innovation. As a subset of innovation, the generalized conclusions regarding the relationship between digital technology and innovation may not be applicable to the specific context of digital technology and breakthrough innovation. Conversely, the debate surrounding the influence of digital technology on innovation may also extend to breakthrough innovation. Furthermore, researchers have realized that digital technology adoption has different dimensions (i.e., depth and breadth, Zhou et al. [15]), which might have different effects on breakthrough innovation. However, empirical research is still lacking in terms of illustrating these effects and their inner mechanisms.

According to the knowledge-based view (KBV), knowledge constitutes an organization’s most valuable asset and is crucial for facilitating breakthrough innovation [16]. Digital technologies can not only foster new methods for knowledge creation through data analysis [12] but also generate new forms of knowledge by enabling complementary insights across different practice fields [17]. Through the spiral transformation of knowledge within the organization, enterprises can continuously create new knowledge to sustain their competitive advantage [18]. Thus, to fill the aforementioned research gaps, this study explores how firms use digital technologies to realize breakthrough innovations from the KBV perspective.

From this perspective, a firm’s innovation capability is affected by how it mobilizes and uses resources related to knowledge [19]. The existing literature suggests that any innovation relies significantly on the recombination of previously existing knowledge [20]. Notably, the combination of novel knowledge components can substantially impact breakthrough innovations compared with the combination of familiar knowledge components [21]. Additionally, digital technology is recognized as a potentially powerful channel for achieving knowledge integration [22]. Consequently, knowledge recombination novelty may serve as a crucial mediating mechanism through which digital technology adoption influences breakthrough innovations.

Furthermore, realizing breakthrough innovations necessitates that firms recombine diverse internal and external knowledge resources [23]. Therefore, breakthrough innovations achieved by a firm through knowledge recombination are inevitably influenced by the interaction between external knowledge brought by digital technology and internal knowledge of the firm. The existing literature suggests that a knowledge base can represent firms’ internal knowledge [16]. Hence, we further consider the moderating effect of the knowledge base on the effectiveness of digital technology adoption.

In summary, the aims of this paper are to answer the following three research questions: (1) How do different dimensions of digital technology adoption affect breakthrough innovations? (2) What are the underlying mechanisms driving this impact? (3) What are the boundary conditions of this impact? For evidence, we use listed Chinese biomedical companies as the context of our study. Our sample comprises 469 listed Chinese biomedical companies covering 5628 firm-year observations from 2010 to 2021. During this period, China’s biomedical industry experienced explosive growth. According to data from Frost & Sullivan (https://www.frost.com (accessed on 16 May 2024)), China’s R&D expenditure in the biomedical sector grew at a compound annual growth rate (CAGR) of 20.15% from 2016 to 2021, far exceeding the global CAGR of 7.42% during the same timeframe, which has led to a series of technological breakthroughs (e.g., immuno- and cell-therapy breakthroughs). Concurrently, the pace of digitalization in China has also accelerated substantially. As stated in a 2022 white paper released by CAICT regarding the development of China’s digital economy, China’s digital economy ranked second only to that of the United States and had the world’s largest digital market. In addition, considering the impact of unobserved unit-specific and time-specific confounders on the estimated causal effects of panel data [24], we employed a two-way fixed-effects regression model that incorporates both unit and time-fixed effects for our empirical analysis.

This study makes contributions in the following three ways: First, it investigates the distinct effects of the two dimensions of digital technology adoption on breakthrough innovations, thereby extending previous research on the antecedents of such innovations. Second, based on KBV, this study reveals the mediating effect of knowledge recombination novelty and the conditional indirect effect of the knowledge base, thus enriching the understanding of the mechanisms and boundary conditions associated with digital technology-driven breakthrough innovation. Third, this research offers practical guidance for business managers regarding the optimal combination of digital technology adoption, knowledge recombination novelty, and knowledge base, enabling enterprises to achieve enhanced levels of breakthrough innovation performance.

The remainder of this paper is organized as follows: Section 2 reviews the literature on digital technology and breakthrough innovation and proposes hypotheses. Section 3 details the methodology, and Section 4 presents the regression and additional analysis results. Section 5 discusses the main findings. Finally, Section 6 concludes with the key insights and presents managerial implications, limitations, and future directions.

2. Theory and Hypotheses

2.1. Digital Technology and Breakthrough Innovation

2.1.1. Inconsistency in Digital Technology’s Effects on Breakthrough Innovation

Digital technology, defined as a combination of information, computing, communication, and connectivity technologies, has already been integrated into numerous business processes [25]. Several studies have examined the influence of digital technology on firms’ innovation activities. At the individual level, the existing literature indicates that certain characteristics of digital technology, such as reliability, portability, and real-time processing, may adversely affect employees’ innovation capabilities [11,26]. Conversely, at the organizational level, current research generally suggests that digital technology positively influences firms’ innovation performance. For example, Ranta et al. [27] found that digital technologies can enhance firms’ resource flows, value creation, and capture, thereby facilitating business model innovation. Additionally, an analysis of manufacturing firms using quantitative data shows that digital technology adoption is positively associated with product innovation [10]. At the geographical area level, the negative aspects of digital technology mainly focus on the digital divide. For instance, Wan et al. [28] used regional panel data to demonstrate digital technology development’s negative spatial spillover effect on innovation performance. Across individual, organization, and geographical area levels, these studies have examined the effect of digital technology on different types of innovation, reflecting that there remains considerable debate about the “digital technology and innovation” relationship. As a subset of innovation, whether there is a similar debate between breakthrough innovation and digital technology is less addressed by existing research.

Although some scholars have investigated the impact of single digital technologies on breakthrough innovation, the existing literature remains inconclusive regarding the overall impact of digital technology adoption on firms’ breakthrough innovations. Several researchers have argued that digital technology positively influences firms’ capability to achieve breakthrough innovations. For instance, a survey conducted by Mikalef et al. [29] indicated that big data analytics can help firms build a competitive advantage by mediating the role of dynamic capabilities. Similarly, based on news data, Johnson et al. [30] discovered that artificial intelligence is primarily utilized for exploratory research and development aimed at fostering breakthrough innovations in novel markets and operational areas rather than enhancing existing product markets and activities.

Conversely, some scholars contend that digital technology may not positively impact firms’ breakthrough innovations and may even have negative effects. The work of Wu et al. [31] is among the earliest studies to examine the impact of digital technology on innovation. They found that although data analysis can enhance process improvement and facilitate the recombination of current technologies, it exerts minimal influence on the creation of novel innovations. That is to say, the adoption of digital technology might increase the number of patents, but it does not necessarily lead to breakthrough innovation. Additionally, empirical evidence suggests that established firms often struggle to adapt to radical technological changes [32], with organizational inertia impeding exploratory innovations [33]. Through their case study of the Polaroid Corporation, Tripsas and Gavetti [32] found that although the company’s scientists embraced digital technology and invented various novel cameras, managers faced difficulty in relinquishing their preconceived notions about acceptable product forms, which hindered the firm’s breakthrough innovations.

In the existing literature, the inconclusive impact of digital technology adoption on firms’ breakthrough innovations can be attributed primarily to previous studies that tend to be limited to the effects of single specific technologies, overlooking the influence of a digital technology portfolio. In practice, a firm’s innovation performance typically results from the interaction of multiple technologies and the external environment [34]. A particular type of digital technology alone is unlikely to alter the external environment for businesses [35]. Furthermore, digital technologies are interrelated and build upon one another because they rely on data generation [10]. Therefore, the effectiveness of digital technology should be evaluated not in isolation but as a cohesive whole.

Given that firms adopt varying investment strategies, they may choose to either increase investments in specific types of digital technologies or invest across a broader array (i.e., more types) of digital technologies [36]. In this context, we divided digital technology adoption into two dimensions: depth and breadth. The depth dimension evaluates the extent to which the full potential of a technology can be used [10]. When a technology is employed to its fullest capacity, a deep knowledge base has been established around it, which, in turn, serves as a valuable resource for breakthrough innovation. Conversely, the breadth dimension indicates whether a technology has been adopted without reflecting the extent of its use [37]. We contend that the greater the variety of digital technologies that a firm uses, the more diverse the knowledge available for breakthrough innovation. Thus, by defining these two dimensions of digital technology adoption (depth and breadth), we can effectively capture a firm’s investment portfolio in digital technologies, thereby evaluating the effect of digital technology on breakthrough innovation from a holistic perspective.

2.1.2. Digital Technology Depth and Breakthrough Innovation

According to KBV, knowledge serves as a core resource for firms and significantly influences breakthrough innovation [16]. Digital technology can transform data into knowledge [38], which facilitates the acquisition and application of knowledge [39]. By effectively leveraging digital technologies to optimize knowledge applications, firms can increase their likelihood of achieving breakthrough innovations. Thus, we propose that digital technology depth positively relates to firms’ breakthrough innovations.

More specifically, in-depth engagement in digital technologies enhances firms’ capacities to process and analyze vast amounts of information from diverse sources [40]. For instance, improving the intensity of data mining enables firms to extract valuable information and knowledge that may not be readily apparent from large volumes of incomplete, noisy, fuzzy, and random practical application data [41]. Additionally, the in-depth utilization of digital technologies facilitates enterprises’ analysis, interpretation, and understanding of the knowledge derived from the data [42]. When confronted with knowledge concealed within complex and large-scale datasets, data visualization technology enables firms to represent knowledge intuitively and effectively through various graphs [43]. Based on the results generated by data visualization, firms can better understand market needs and carry out product innovations. Moreover, with a more profound use of digital technologies, enterprises integrate the acquired new knowledge with existing knowledge more efficiently and develop concrete applications or products based on it [44]. For example, enterprises’ construction of digital platforms through digital technologies can promote knowledge sharing between internal and external actors. Based on internalized new knowledge, enterprises can improve product innovation probability [45].

Overall, as the use of digital technologies increases, firms are better equipped to process, analyze, and leverage knowledge embedded in data, enabling them to uncover hidden market opportunities and implement ideas from emerging technological and market domains. Therefore, we propose the following:

H1a. 

Digital technology depth is positively related to firms’ breakthrough innovations.

2.1.3. Digital Technology Breadth and Breakthrough Innovation

Firms can expand access to knowledge to increase the likelihood of breakthrough innovations by adopting various digital technologies (e.g., artificial intelligence, blockchain, big data, etc.) as well. Consequently, we propose that the breadth of digital technology is positively correlated with firms’ breakthrough innovations.

More specifically, as the number of digital technologies adopted increases, enterprises can expand the data search scope and obtain a broader range of knowledge sources [46]. Data search across technology types and organizational boundaries assists enterprises in introducing new elements and generating major innovative technologies [47]. Additionally, adopting multiple digital technologies helps firms enhance knowledge accumulation and acquire exclusive knowledge [48], for example, by combining the Internet of Things (IoT), cloud computing (CC), and predictive analytics (PA). The fundamental functions of IoT (e.g., sensing, connectivity) are responsible for the collection and transmission of user data [49]; CC is responsible for creating large databases and deploying applications that access, process, and share data [50], and PA is responsible for analyzing user data to create new knowledge [38]. During this process, with the accumulation of knowledge, firms are better positioned to capture the potential value, thus improving the business model [51]. Simultaneously, enterprises accumulate heterogeneous knowledge through their knowledge base and utilize various digital technologies to establish a platform for sharing technical resources and knowledge. This facilitates cooperation and mutual learning among personnel from different functional departments within the enterprise, thereby stimulating new ideas [52]. Therefore, we propose the following:

H1b. 

Digital technology breadth is positively related to firms’ breakthrough innovations.

2.2. Knowledge Recombination Novelty

In the knowledge recombination literature, scholars tend to characterize knowledge as discrete elements or components [53,54]. Knowledge recombination novelty reflects the degree to which an invention is original in its recombination of components and principles to achieve its intended purpose [55]. Given that radical inventions are viewed as combining existing or new components in unprecedented ways [56], improving the novelty of knowledge recombination is essential for firms seeking breakthrough innovations. Moreover, because the different impacts of digital technology depth and breadth on breakthrough innovation depend on knowledge-recombinant novelty, employing knowledge-recombinant novelty as a mediator can enhance the understanding of the relationship between digital technology and breakthrough innovation.

Firstly, from an overall perspective, we posit that a firm’s digital technology portfolio is directly related to its level of knowledge recombination novelty. Regarding digital technology depth, first, knowledge recombination novelty necessitates that firms discern unexplored interdependencies among technologies in a “capability broadening” exercise [57]. Digital technology depth provides a channel through which firms can fulfill this capability-breaking requirement. Specifically, digital technology can partition internal and external knowledge firms obtain into visible design rules precisely, unambiguously, and completely [58]. To firms, these fully specified and standardized component interfaces are conducive to the later recombination of modules. Second, digital technology depth aids enterprises in resolving cognitive discrepancies [59], identifying and selecting the most compatible and valuable knowledge for recombination [20], and forming new combinations of previously unimaginable knowledge. For example, during the new product development process, the addition of digital technology can replace elements of the brainwork that have been widely employed in conventional production techniques [60]. With the assistance of digital technology, researchers are able to model and simulate products, better comprehend product characteristics, and develop the best product portfolio to match market expectations [61].

Regarding the breadth of digital technology, first, the scope of knowledge for recombination is crucial to the success of this process [62]. Digital technology breadth reflects the extent of a firm’s digital technology portfolio [37]. Consequently, the more types of digital technologies a firm adopts, the more it can broaden its search for both internal and external knowledge, thus promoting the recombination of new technology combinations across a wider array of domains. Second, the breadth of digital technology facilitates companies in building standardized platforms that enable inventors within the company to understand different domains of specialized knowledge. Digital technology can encode the different types of knowledge in a unified manner; thus, knowledge can be understood as processed data in essence. These processed data enable inventors to comprehend the different types of knowledge and, thus, recombine them in new ways [63]. In addition, as heterogeneous knowledge increases, the probability of the emergence of new technology combinations also increases. From this perspective, the digital technology breadth is likely to enhance firms’ knowledge recombination novelty.

Furthermore, with the improvement in the depth and breadth of digital technology, firms can expand the depth and scope of their knowledge search, thereby enriching their knowledge storage. According to knowledge management theory, combining old and new elements stimulates firms’ technological concepts and creative capabilities, ultimately leading to the development of breakthrough technologies [64]. Thus, we propose that knowledge recombinant novelty can enhance firms’ breakthrough innovations.

More specifically, recombination creation can yield more innovative and unique combinations of knowledge, thereby enriching firms’ knowledge bases. According to resource-based theory, the knowledge base is the most critical resource for sustainable breakthrough innovations in firms [65]. Consequently, a higher level of novelty in knowledge recombination increases the likelihood that a firm will develop a unique knowledge base essential for achieving breakthrough innovations. Additionally, utilizing existing components in new combinations presents opportunities for entirely novel inventions [56]. In fact, many breakthrough inventions of the past were created by combining existing but previously disparate components [66]. For instance, the computerized axial tomography scanner, which combined X-ray imaging with distinct knowledge domains in computer science and mathematics, demonstrated exceptional 3D reconstruction capabilities. This innovation has inspired a range of advancements in imaging technologies, such as positron emission tomography, computerized tomography simulation, and industrial computerized tomography [67]. Moreover, recombining familiar components in novel ways enables firms to generate breakthrough inventions [68,69]. A notable instance is Nobel Prize winner Kary Mullis’s invention of polymerase chain reaction, a biomedical technique used to identify and amplify DNA sequences that have significantly advanced progress in biotechnology.

Overall, knowledge recombination novelty improves a firm’s capability for breakthrough innovations in two primary ways: (1) by using existing components in new combinations and (2) by recombining familiar components in novel ways.

Accordingly, we propose the following:

H2a. 

Knowledge-recombinant novelty mediates the relationship between digital technology depth and breakthrough innovations;

H2b. 

Knowledge-recombinant novelty mediates the relationship between digital technology breadth and breakthrough innovations.

2.3. Knowledge Base

According to KBV, a firm’s existing knowledge base delineates its scope and capacity to understand and utilize new knowledge for breakthrough innovations [70]. In organizing breakthrough innovations, firms face a choice between internal and external knowledge-development approaches [71]. The internal mode involves enterprises assembling internal knowledge to generate innovations, while the external mode involves integrating external knowledge through innovation networks [72]. Regarding internal knowledge development, the knowledge base reflects the structure and content of a firm’s existing knowledge, serving as its most unique resource for breakthrough innovations [73]. The knowledge base can be categorized into two dimensions: knowledge breadth, which indicates the range of distinct and multiple domains contained within the firm’s knowledge repository, and knowledge depth, which pertains to the sophistication and complexity of knowledge in critical fields [16]. Regarding external knowledge acquisition, digital technology, as a new means for enterprises to obtain external knowledge, has changed the way in which knowledge is created within an organization and has a notably positive effect on knowledge creation processes, particularly knowledge recombination [74,75].

To handle all the necessary development required for breakthroughs, enterprises often need to recombine internally and externally sourced knowledge [76]. In other words, the development of internal knowledge and the acquisition of external knowledge tend to interact during the recombination process. Therefore, the knowledge base can influence the indirect relationship between digital technology and breakthrough innovation by shaping the impact of digital technology on knowledge recombination novelty.

First, we propose that the level of a firm’s knowledge base influences the positive relationship between digital technology depth and knowledge recombinant novelty. On the one hand, acquiring depth in knowledge within a specific domain increases the number of knowledge attributes, thereby enhancing the potential for recombinations [77]. In addition, possessing deep knowledge in a particular area enables firms to use their knowledge more effectively, allowing them to identify and select new linkages that are more promising for developing novel and valuable outcomes [78]. However, the performance gains of any technological method exhibit a characteristic of diminishing returns [1]. In other words, as the level of knowledge depth within enterprises increases, the impact of digital technology depth on knowledge recombination novelty diminishes. On the other hand, considering the complementarity between digital technology depth and knowledge breadth, digital technology depth is conducive to increasing the complexity of a firm’s knowledge structure. When a firm possesses a high level of knowledge breadth, digital technology depth facilitates the establishment of the necessary building blocks to activate the recombination process that generates creative ideas [77]. Further, digital technology depth empowers enterprises to assess the novelty of knowledge recombination.

Second, we propose that the level of a firm’s knowledge base influences the positive relationship between digital technology breadth and knowledge recombinant novelty. On the one hand, knowledge recombinant novelty necessitates organizations to sustain a certain extent of breadth in their knowledge—specifically, knowledge across multiple technological disciplines. When an enterprise has developed deep knowledge and core competencies, which are manifested as technical or professional expertise, it tends to focus on activities within existing areas of expertise [79]. However, while boosting the creativity of the firm, knowledge depth in the specialized domain can also have a detrimental effect on the firm’s creativity. Specifically, knowledge depth can lead to cognitive rigidity, restricting a firm’s ability to adapt its knowledge structure to generate novel combinations [80]. Digital technology breadth provides firms with access to diverse knowledge domains. Exposure to new knowledge from different areas enables enterprises to refresh their knowledge base, break the existing cognitive structure, and improve the probability of breaking through the existing organizational process and routine to find new combinations [78]. Consequently, when the level of knowledge depth within enterprises is high, the benefits derived from digital technology breadth in terms of knowledge recombination novelty will be amplified.

On the other hand, similar to the relationship between digital technology depth and knowledge depth, digital technology breadth exhibits diminishing returns regarding knowledge recombination novelty. When enterprises possess information about heterogeneous market segments, the marginal benefit of acquiring additional market knowledge to generate breakthrough ideas diminishes [16]. In other words, when an enterprise has a high level of knowledge breadth, the overlapping information brought about by the breadth of digital technology may make a minor refinement or extension of the existing knowledge base. However, it does not improve the recombination novelty of the enterprise.

Furthermore, we propose that the knowledge base serves as a moderator between the depth (breadth) of digital technology, knowledge recombination novelty, and breakthrough innovations. On the one hand, previous studies have indicated that the indirect relationship begins with the depth (breadth) of digital technology, which then influences knowledge recombination novelty and ultimately leads to breakthrough innovations. Considering that the positive relationship between digital technology depth (digital technology breadth) and knowledge recombination novelty is affected by the level of a firm’s knowledge base, it is expected that the knowledge base will also affect this indirect relationship. On the other hand, from the perspective of resource-based theory concerning information technology, it is essential to consider contingency factors when analyzing the impact of information technology competency on firms’ strategies [81]. Ko and Liu [73] found that the knowledge base, as a source of internal knowledge in firms, can serve as a contingency factor that complements firms’ information technology competency in fostering innovation. Given that digital technology is a subset of information technology [25], the knowledge base can serve as a contingency factor that influences the indirect relationship between digital technology adoption and breakthrough innovations by affecting how digital technology adoption impacts knowledge recombination novelty. Therefore, we propose the following hypotheses:

H3a₁. 

The positive relationship between digital technology depth and knowledge recombination novelty diminishes as the level of a firm’s knowledge base increases, and the knowledge base suppresses the mediating effect of knowledge recombination novelty between digital technology depth and breakthrough innovations;

H3a₂. 

The positive relationship between digital technology depth and knowledge recombination novelty strengthens as the level of a firm’s knowledge base increases, and the knowledge base improves the mediating effect of knowledge recombination novelty between digital technology depth and breakthrough innovations;

H3b₁. 

The positive relationship between digital technology breadth and knowledge recombination novelty strengthens as the level of a firm’s knowledge base increases, and the knowledge base improves the mediating effect of knowledge recombination novelty between digital technology breadth and breakthrough innovations;

H3b₂. 

The positive relationship between digital technology breadth and knowledge recombination novelty diminishes as the level of a firm’s knowledge base increases, and the knowledge base suppresses the mediating effect of knowledge recombination novelty between digital technology breadth and breakthrough innovations.

Our conceptual model and hypotheses are illustrated in Figure 1.

3. Methodology

3.1. Data

This study selected Chinese biomedical listed companies as the sample for testing our conceptual model for several reasons. First, digital technology is extensively utilized in China. The “2024 Global Unicorn Ranking (https://www.hurun.net (accessed on 7 July 2024))”, released by the Hurun Research Institute, indicates that China has 340 companies on the list, ranking second globally. Many of these companies thrive due to advancements in digital technology. Second, biotechnology is an interdisciplinary field with a rich history of R&D. Its evolution is driven by the integration of existing yet disparate knowledge and technologies [82]. Furthermore, biotechnology has seen numerous breakthroughs over the years, with the vast majority of technological inventions being patented [21], providing ample opportunities to validate our hypotheses. Third, unlike other domestic firms, Chinese listed companies are governed by a more stringent regulatory system and are required to provide more comprehensive data disclosures [35]. Therefore, this study focuses on biomedical companies listed in China, which facilitates the construction of a comprehensive explanatory framework for understanding how to achieve breakthrough innovations driven by digital technology.

To obtain the panel data set of Chinese biomedical listed companies, we first acquired the directory of 5361 companies listed on the A-stock market of the Shanghai and Shenzhen Stock Exchange from the China Stock Market and Accounting Research (CSMAR) (The CSMAR database is designed and developed by Shenzhen CSMAR Data Technology Co., Ltd. (Shenzhen, China). The data cover the fields of economy, listed companies, stocks, funds, bonds, stock index futures, commodity futures, and Hong Kong stocks (see https://www.csmar.com/index.html (accessed on 3 April 2024)).) database. Subsequently, we filtered the listed companies to identify those belonging to the biomedical industry, using the Directory of Industrial Classification of ShenYin and WanGuo. After the first screening round, our initial sample contained 480 listed companies.

We utilized four sub-databases in the CSMAR database to obtain the data. First, we gathered information on firms’ adoption of digital technology from the China Listed Firm’s Digital Transformation Research (EDT) database. Second, we acquired breakthrough innovation and knowledge recombination novelty information from the Patent Cited Research database and the China Listed Firm’s R&D Innovation Research (LCPT) database. Third, we collected information on control variables from the LCPT database and the China Stock Market Financial Statements Derivative Research database. These four sub-databases contain comprehensive information on relevant research topics and fields from authoritative sources, making them superior to other similar domestic databases in terms of completeness, accuracy, and timeliness.

Given that companies with significant data missing may adversely affect our forecast results, we conducted a second round of screening of the initial sample to ensure the rigor and standardization of the empirical analysis. Specifically, we (1) excluded ST and *ST companies, (2) excluded companies with a severe lack of financial and accounting data, and (3) excluded companies with mismatched International Patent Classification (IPC) numbers and patent citation data or those with substantial missing information. After removing companies with incomplete data, the final panel dataset comprised 469 listed companies and 5628 firm-year observations from 2010 to 2021.

3.2. Measurement

3.2.1. Dependent Variable: Breakthrough Innovation

To measure firms’ breakthrough innovation performance, we utilized patents, which are one of the primary outcomes of a firm’s R&D efforts. By definition, patents must describe inventions that are novel and non-obvious. Specifically, an invention is eligible for patenting only if it has not been previously disclosed in prior publications (novel) and represents a non-trivial advancement in the state of the art (non-obvious) [83]. Furthermore, the information contained in patent documents can reflect patent quality or value after screening. Several studies have demonstrated a strong relationship between patent quality or value and the assessment and characterization of breakthrough innovation [84]. Therefore, patents serve as a relatively objective measure of breakthrough innovation.

To define firms’ breakthrough innovations, we adhered to established practices in the literature and measured them by the number of patent citations and the count of invention patents [85,86]. Specifically, we used three indicators: forward citations of patents, top 10% citations, and the number of invention patents to define breakthrough innovation.

Forward citation of patents (Citation). The volume of citations attributed to a patent is closely linked to its significance within its technical field, serving as a crucial criterion for breakthrough innovation [87]. Citation acts as a direct indicator of the impact of the underlying technology, illustrating how a patent contributes to subsequent technological advancements. Ahuja and Morris Lampert [3] noted that breakthrough innovation represents a form of leapfrogging and non-linear innovation, initiating a series of subsequent innovations by establishing a new technological trajectory. As the original patent, the number of citations for breakthrough innovation reflects the essential characteristics of its pioneering contribution. In summary, a higher number of citations correlates with the greater importance of the patent in its technological field and a more substantial contribution to subsequent innovations.

Due to the varying degrees of truncation associated with patents of different ages, recently granted patents typically receive fewer citations compared to their older counterparts. Accounting for this truncation error, we calculated the ratio of citations obtained by the patents owned by the focal firm to the average citations within the industry for that specific year [35]. Subsequently, we employed the logarithm of this ratio, plus one, as a measure of the firm’s level of breakthrough innovation.

Top 10% citations (Ci90). We defined Ci90 as the proportion of the top 10% of most cited patents filed by a firm in a given year relative to the total number of patents filed by that firm in the same year [85]. The top 10% of most cited patents were identified based on the number of citations received by all patents filed within the same category during that year under review. The more patents with higher citation rankings, the more breakthrough innovations there are. Additionally, we addressed the truncation issue related to patent age by employing the same method used in the analysis of forward citations.

Number of invention patents (Invenum). Regarding features, the patents were classified into three types: invention, utility model, and design [88]. Invention patents are defined as novel technical proposals for products or methods [89]. Compared to utility model patents and design patents, they are regarded as more accurate indicators of an enterprise’s breakthrough innovation capabilities and performance [90]. Consequently, we utilized the logarithm of the number of invention patents plus one as the third measure of breakthrough innovation [35].

3.2.2. Independent Variables: Digital Technology Depth and Digital Technology Breadth

Digital technology adoption is hardly measurable by a single financial indicator [10]. The significance that a firm attaches to a particular strategic direction can be reflected in the frequency of associated keywords appearing in its annual report [91]. Therefore, the current research primarily employs textual analysis to assess the adoption of digital technologies by constructing a dictionary of relevant terms and counting their occurrences in firms’ financial reports as a proxy for the extent of digital activities [15]. In line with these processing principles, the technology-driven section of the EDT database presents a detailed statistical table indicating the word frequency of digital technologies, which encompasses four categories: artificial intelligence, blockchain, cloud computing, and big data. Based on the digital technology types and corresponding word frequency information reported by the sample companies in the statistical table of the word frequency of digital technologies, referring to Yang et al. [37], we measured digital technology depth (DTD) as the ratio of the total number of digital technology keywords to the number of digital technology types reported by the company, and digital technology breadth (DTB) was measured by the number of digital technology types reported by the company. The specific calculation formulas are as follows:

$$DT{D}_{i,t}={\displaystyle \frac{{\displaystyle {\sum }_j{n}_{i,j,t}}}{{\displaystyle {\sum }_{j}D{T}_{i,j,t}}}}$$

(1)

$$DT{B}_{i,t}={\displaystyle {\sum }_{j}D{T}_{i,j,t}}$$

(2)

where DTD_i,t and DTB_i,t denote the depth and breadth of digital technology (DT) of the company i in year t, respectively; DT_i,j,t is a binary variable, with DT_i,j,t = 1 if the j-th DT is mentioned at least once in company i’s annual report for year t, 0 otherwise, and n_i,j,t refers to the number of relevant keywords associated with the j-th DT that are included in company i’s annual report for year t.

3.2.3. Mediator Variable: Knowledge Recombination Novelty

To operationalize knowledge recombination novelty (KRN), we used IPC numbers to go through all subclass assignments of all World Intellectual Property Organization (WIPO) granted patents to identify all previously uncombined pairs of subclasses at the level of a single patent [55]. To construct this variable, we first created a list of IPC code dyads containing all unique subclass pairs for each patent based on the biomedical patents of the sample companies. Second, we compared these subclass pairs with all patent subclass pairs applied by all sample companies in the past five years to determine which subclass pairs were first-time occurrences. Finally, to normalize our measure, we divided the total number of subclass pairs appearing for the first time by the total number of subclass pairs in the corresponding companies’ patent databases. The detailed formula for calculation is presented below:

$$KR{N}_{i,t}={\displaystyle \frac{TFS{P}_{i,t}}{TS{P}_{i,t}}}$$

(3)

where KRN_i,t denotes the novelty of knowledge recombination of the company i in year t; TFSP_i,t refers to the total number of first-appearing subclass pairs (TFSP) owned by the company i in all patent subclass pairs in year t, and TSP_i,t represents the total number of all patent subclass pairs (TSP) for the company i in year t.

3.2.4. Moderator Variable: Knowledge Base

Following Ko and Liu [73], the knowledge base (KB) comprises two dimensions: knowledge breadth and knowledge depth. Given that the information regarding the technology category to which a patent belongs is the most feasible means of operationalizing both the breadth and depth of knowledge [92], we first processed the biomedical patent data of the sample companies to construct KB. Specifically, we extracted the portion preceding the “/” in the IPC number of each patent as a representative of the knowledge element [64]. Second, we used the information on each patent’s IPC number to measure knowledge breadth and depth. Concretely, since IPC numbers represent different types of knowledge, we measured knowledge breadth as the number of different IPC numbers that a firm has in the observation year, and knowledge depth was measured by the ratio of the total number of IPC numbers owned by a firm in the observation year to the number of different IPC numbers owned by a firm. Finally, to avoid the subjectivity associated with determining indicator weights and to reduce estimation result deviations arising from information overlap among statistical indicators, we adopted the entropy method to synthesize the two indicators of knowledge breadth and knowledge depth to calculate KB.

3.2.5. Control Variables

Previous research has indicated that specific firm characteristics could influence breakthrough innovation [3,93]. Consequently, we introduced firm size, leverage ratio, return on assets, growth, cash flow, and R&D intensity as control variables in this research. Firm size (Size) was measured using the firm’s total assets and standardized accordingly [94]. The leverage ratio (Lev) was calculated as the ratio of total liabilities to total assets at the end of the year [35]. Return on assets (ROA) was calculated using the net profit ratio to total assets [95]. Growth (Growth) was measured at the annual revenue growth rate [20]. Cash flow (Cashflow) was determined by calculating net cash flows from operating activities as a ratio of total assets [35]. Finally, we defined R&D intensity (RD) as the ratio of R&D expenditures to total sales [96].

Table 1. Descriptive statistics.

Variable	Observation	Mean	Standard Deviation	Minimum	Maximum
Citation	5628	0.3490	0.6373	0.0000	4.6258
Ci90	5628	0.3568	0.6766	0.0000	3.3212
Invenum	5628	0.8860	1.6680	0.0000	8.8432
DTD	5628	0.3611	2.5634	0.0000	94.0000
DTB	5628	0.1544	0.4921	0.0000	4.0000
KRN	5628	0.1605	0.2677	0.0000	1.0000
KB	5628	0.0279	0.0476	0.0000	0.5496
Size	5628	0.0000	1.0000	−0.3983	20.3916
Lev	5628	0.1903	0.3196	0.0000	12.1274
ROA	5628	0.0322	0.0870	−2.4911	0.6042
Growth	5628	0.1196	0.6301	−0.9484	19.7651
Cashflow	5628	0.0203	0.0636	−1.8234	0.6737
RD	5628	0.4298	8.7387	0.0000	317.2884

presents the descriptive statistics of the variables.

3.3. Regression Model

We implemented a one-year time lag between the adoption of digital technology and breakthrough innovations, as the adoption of digital technology by firms for breakthrough innovations typically exhibits a lagged effect. This approach was adopted to mitigate reverse causality, in line with the empirical setup established by Dong and Yang [20] and Liu et al. [35]. Furthermore, we utilized fixed-effect models with firm and year-fixed effects according to prior research [97]. First, we established the following baseline models to test the proposed hypotheses H1a and H1b:

$$B{I}_{i,t+1}={\beta }_0+{\beta }_{1}DT{D}_{i,t}+{\beta }_{2}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(4)

$$B{I}_{i,t+1}={\beta }_0+{\beta }_{1}DT{B}_{i,t}+{\beta }_{2}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(5)

where i represents the firm; t denotes the year, and BI_t+1 indicates the firm’s breakthrough innovation (BI) performance during period t + 1. We evaluated breakthrough innovation through three distinct measures, as outlined in Section 3.2.1. DTD in Equation (4) indicates digital technology depth, and DTB in Equation (5) indicates digital technology breadth. Controls include the control variables discussed above: Size, ROA, Lev, Growth, Cashflow, and RD. φ_i and γ_t denote firm and year-fixed effects, respectively. ε_i,t represents a random error term.

Second, we used Equations (6) to (9) to verify H2a and H2b, following Baron and Kenny [98]. Specifically, Equations (6) and (7) were utilized to examine the mediating effect of KRN between DTD and BI, and Equations (8) and (9) were used to assess the mediating effect of KRN between DTB and BI. Additionally, knowledge recombination novelty in the subsequent year was used as the mediating variable.

$$KR{N}_{i,t+1}={\beta }_0+{\beta }_{1}DT{D}_{i,t}+{\beta }_{2}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(6)

$$B{I}_{i,t+1}={\beta }_0+{\beta }_{1}DT{D}_{i,t}+{\beta }_{2}KR{N}_{i,t+1}+{\beta }_{3}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(7)

$$KR{N}_{i,t+1}={\beta }_0+{\beta }_{1}DT{B}_{i,t}+{\beta }_{2}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(8)

$$B{I}_{i,t+1}={\beta }_0+{\beta }_{1}DT{B}_{i,t}+{\beta }_{2}KR{N}_{i,t+1}+{\beta }_{3}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(9)

Finally, based on Equations (6) to (9), we used the knowledge base of the subsequent year as the moderating variable and added the interaction terms of the knowledge base and independent variables to test H3a₁, H3a₂, H3b₁, and H3b₂, in accordance with Hayes and Rockwood [99].

$$\begin{array}{cc}KR{N}_{i,t+1}& ={\beta }_0+{\beta }_{1}DT{D}_{i,t}+{\beta }_{2}K{B}_{i,t+1}+{\beta }_{3}DT{D}_{i,t}\times K{B}_{i,t+1}\hfill \\ & +{\beta }_{4}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}\hfill \end{array}$$

(10)

$$\begin{array}{cc}B{I}_{i,t+1}& ={\beta }_0+{\beta }_{1}DT{D}_{i,t}+{\beta }_{2}KR{N}_{i,t+1}+{\beta }_{3}K{B}_{i,t+1}+{\beta }_{4}DT{D}_{i,t}\times K{B}_{i,t+1}\hfill \\ & +{\beta }_{5}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}\hfill \end{array}$$

(11)

$$\begin{array}{cc}KR{N}_{i,t+1}& ={\beta }_0+{\beta }_{1}DT{B}_{i,t}+{\beta }_{2}K{B}_{i,t+1}+{\beta }_{3}DT{B}_{i,t}\times K{B}_{i,t+1}\hfill \\ & +{\beta }_{4}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}\hfill \end{array}$$

(12)

$$\begin{array}{cc}B{I}_{i,t+1}& ={\beta }_0+{\beta }_{1}DT{B}_{i,t}+{\beta }_{2}KR{N}_{i,t+1}+{\beta }_{3}K{B}_{i,t+1}+{\beta }_{4}DT{B}_{i,t}\times K{B}_{i,t+1}\hfill \\ & +{\beta }_{5}Control{s}_{i,t}+{\phi }_i+{\gamma }_t+{\epsilon }_{i,t}\hfill \end{array}$$

(13)

In Equations (10) and (11), the item $({\beta }_1+{\beta }_{3}K{B}_{i,t+1}){\beta }_2$ denotes the degree to which the indirect effect of DTD on BI through KRN is affected by KB. In Equations (12) and (13), the item $({\beta }_1+{\beta }_{3}K{B}_{i,t+1}){\beta }_2$ denotes the degree to which the indirect effect of DTB on BI through KRN is influenced by KB. We utilized Stata 15.1 software to establish the aforementioned models.

4. Results

4.1. Regression Analysis

Results are presented in

Table 2. The effects of digital technology adoption on firms’ breakthrough innovations.

Variables	Citation		Ci90		Invenum
	(1)	(2)	(3)	(4)	(5)	(6)
Size	−0.0292	−0.0365	−0.1258 ***	−0.1347 ***	−0.4188 ***	−0.4449 ***
	(−0.8851)	(−1.0873)	(−4.0136)	(−4.2004)	(−4.3217)	(−4.5776)
ROA	−0.0580 *	−0.0566	0.2130 ***	0.2140 ***	0.5939 **	0.5948 **
	(−1.6598)	(−1.5966)	(3.2993)	(3.2981)	(2.0434)	(2.0606)
Lev	0.0358	0.0382	0.0620 **	0.0646 **	0.0384	0.0454
	(1.6147)	(1.6172)	(2.0219)	(2.0330)	(0.5125)	(0.6174)
Growth	0.0118 **	0.0110 **	0.0075	0.0065	0.0364	0.0336
	(2.2489)	(2.1516)	(0.4318)	(0.3725)	(1.1711)	(1.0918)
Cashflow	0.3788 ***	0.3508 ***	0.6719 ***	0.6391 **	2.2915 ***	2.1979 ***
	(3.0473)	(2.8092)	(2.6567)	(2.5525)	(3.1695)	(3.0728)
RD	−0.0002 *	−0.0002 **	−0.0000	−0.0000	0.0049	0.0048
	(−1.8407)	(−2.1106)	(−0.0375)	(−0.1020)	(0.9750)	(0.9787)
DTD	0.0119 ***		0.0158 ***		0.0501 ***
	(3.4944)		(3.0750)		(3.7648)
DTB		0.0610 **		0.0736 **		0.2162 **
		(2.5770)		(2.4077)		(2.3336)
Constant	0.2549 ***	0.2540 ***	0.2288 ***	0.2277 ***	0.3184 ***	0.3155 ***
	(17.9058)	(17.7031)	(8.1482)	(8.0911)	(6.0898)	(5.9885)
Year	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5159	5159	5159	5159	5159	5159
R2	0.0475	0.0453	0.0282	0.0270	0.1667	0.1639
Largest VIF	1.1110	1.1110	1.1110	1.1110	1.1110	1.1110

Notes. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

,

Table 3. The mediating effect of knowledge recombination novelty.

Variables	KRN		Citation		Ci90		Invenum
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Size	−0.0820 ***	−0.0843 ***	−0.0213	−0.0284	−0.0573 **	−0.0642 **	−0.3352 ***	−0.3585 ***
	(−4.8952)	(−4.8859)	(−0.6512)	(−0.8530)	(−2.3162)	(−2.5666)	(−3.6489)	(−3.9276)
ROA	0.1345 ***	0.1350 ***	−0.0709 **	−0.0696 *	0.1006 *	0.1010 *	0.4567	0.4563 *
	(4.2937)	(4.2891)	(−2.0292)	(−1.9631)	(1.7935)	(1.7932)	(1.6448)	(1.6568)
Lev	0.0483	0.0491	0.0311	0.0335	0.0216	0.0235	−0.0108	−0.0050
	(1.5831)	(1.5839)	(1.5924)	(1.5948)	(1.1386)	(1.2295)	(−0.1099)	(−0.0517)
Growth	−0.0020	−0.0023	0.0120 **	0.0112 **	0.0092	0.0084	0.0385	0.0360
	(−0.4878)	(−0.5544)	(2.2855)	(2.1933)	(0.5557)	(0.5083)	(1.2860)	(1.2158)
Cashflow	0.3807 ***	0.3717 ***	0.3423 ***	0.3148 **	0.3538	0.3280	1.9035 ***	1.8167 ***
	(4.5037)	(4.4201)	(2.8024)	(2.5583)	(1.4705)	(1.3686)	(2.7404)	(2.6389)
RD	0.0006 ***	0.0006 ***	−0.0003 **	−0.0003 ***	−0.0005 **	−0.0006 **	0.0042	0.0041
	(4.2516)	(4.1700)	(−2.3341)	(−2.6723)	(−2.4242)	(−2.5655)	(0.8456)	(0.8455)
DTD	0.0037 ***		0.0115 ***		0.0127 **		0.0463 ***
	(2.6744)		(3.4604)		(2.5179)		(3.6419)
DTB		0.0194 *		0.0591 **		0.0574 **		0.1963 **
		(1.9355)		(2.5394)		(2.0085)		(2.1932)
KRN			0.0959 ***	0.0967 ***	0.8355 ***	0.8370 ***	1.0193 ***	1.0256 ***
			(3.6831)	(3.7446)	(13.6341)	(13.6857)	(8.3885)	(8.4415)
Constant	0.1364 ***	0.1361 ***	0.2419 ***	0.2408 ***	0.1148 ***	0.1138 ***	0.1794 ***	0.1760 ***
	(12.8986)	(12.7958)	(17.4159)	(17.1425)	(4.1689)	(4.1318)	(3.1601)	(3.0678)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5159	5159	5159	5159	5159	5159	5159	5159
R2	0.1208	0.1206	0.0538	0.0517	0.1159	0.1150	0.1894	0.1869
Largest VIF	1.1110	1.1110	1.1278	1.1278	1.1278	1.1278	1.1278	1.1278

Notes. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

, and

Table 4. The conditional indirect effect of knowledge base.

Variables	KRN		Citation		Ci90		Invenum
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Size	−0.0653 ***	−0.0624 ***	−0.0223 ***	−0.0325 ***	−0.0575 ***	−0.0564 ***	−0.3368 ***	−0.3480 ***
	(−5.4186)	(−5.3439)	(−2.6427)	(−3.8173)	(−3.0134)	(−2.9174)	(−3.8311)	(−4.0182)
ROA	0.0735 ***	0.0733 ***	−0.0877 ***	−0.0882 ***	0.0536	0.0556	0.3659	0.3610 **
	(3.0871)	(3.0946)	(−2.8611)	(−2.8828)	(0.7731)	(0.8006)	(1.3634)	(2.2364)
Lev	0.0364	0.0370	0.0304 **	0.0297 **	0.0189	0.0223	−0.0157	0.0028
	(1.3271)	(1.3376)	(2.2304)	(2.1796)	(0.6135)	(0.7208)	(−0.1735)	(0.0387)
Growth	−0.0033	−0.0035	0.0112 **	0.0114 **	0.0068	0.0057	0.0338	0.0342
	(−0.8980)	(−0.9557)	(2.0538)	(2.0870)	(0.5470)	(0.4594)	(1.1342)	(1.1868)
Cashflow	0.2389 ***	0.2369 ***	0.3052 ***	0.2927 ***	0.2708	0.2480	1.7333 **	1.9692 ***
	(3.2282)	(3.2409)	(4.1872)	(4.0151)	(1.6424)	(1.4999)	(2.5459)	(5.1480)
RD	0.0005 ***	0.0005 ***	−0.0003	−0.0003	−0.0005	−0.0006	0.0042	0.0044 **
	(5.3470)	(5.1766)	(−0.7420)	(−0.6964)	(−0.5796)	(−0.6349)	(0.8510)	(2.1184)
DTD	0.0023 *		0.0041 **		0.0192 ***		0.0460 ***
	(1.7940)		(2.0060)		(4.1094)		(2.9507)
DTB		0.0189 *		−0.0063		0.0863 ***		0.1793 ***
		(1.7482)		(−0.5338)		(3.2345)		(2.8933)
KRN			−0.0249	−0.0193	0.5453 ***	0.5428 ***	0.4363 ***	0.5397 ***
			(−1.3182)	(−1.0228)	(12.7763)	(12.6966)	(3.2182)	(5.4797)
KB	3.1450 ***	3.2421 ***	1.7752 ***	1.5581 ***	4.3614 ***	4.5595 ***	8.7087 ***	9.1922 ***
	(10.4118)	(9.8639)	(13.7154)	(11.5625)	(14.9010)	(14.9209)	(6.2744)	(12.9403)
DTD ✕ KB	−0.0269 ***		0.0568 ***		−0.1117 ***		−0.0887 *
	(−4.7585)		(4.0424)		(−3.5165)		(−1.6772)
DTB ✕ KB		−0.4510 **		0.9900 ***		−1.0605 ***		−1.6359 **
		(−2.5142)		(6.8795)		(−3.2498)		(−2.1635)
Constant	0.0856 ***	0.0840 ***	0.2296 ***	0.2320 ***	0.0841 ***	0.0805 ***	0.1184 **	0.1208 **
	(7.9254)	(7.6579)	(20.0976)	(20.3290)	(3.2544)	(3.1105)	(2.0418)	(2.0064)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5159	5159	5159	5159	5159	5159	5159	5159
R2	0.3060	0.3076	0.0965	0.1005	0.1566	0.1556	0.2173	0.2051
Largest VIF	2.1340	1.2329	2.1422	1.8184	2.1422	1.8184	2.1422	1.8184

Notes. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

, respectively, where breakthrough innovation was measured using Citation, Ci90, and Invenum. We assessed multicollinearity by calculating the variance inflation factors (VIFs) for all predictors in each column. The maximum value of VIF was 2.1422, which is far lower than the threshold of 10.0, suggesting that there is no apparent multicollinearity between the predicted variables [100]. First, we tested the impacts of digital technology adoption on firms’ breakthrough innovations in Table 2. Column (1), Column (3), and Column (5) demonstrated a positive correlation between digital technology depth and firms’ breakthrough innovations (β = 0.0119, β = 0.0158, β = 0.0501; all ps < 0.01). Similarly, Column (2), Column (4), and Column (6) revealed a positive correlation between digital technology breadth and firms’ breakthrough innovations (β = 0.0610, β = 0.0736, β = 0.2162; all ps < 0.05). These results support H1a and H1b, indicating that both dimensions of digital technology adoption positively influence breakthrough innovations.

Second, we examined the mediating effect of knowledge recombination novelty in Table 3. The results presented in Column (1) demonstrate that digital technology depth positively impacts knowledge recombination novelty (β = 0.0037, p < 0.01). Additionally, the results from Column (3), Column (5), and Column (7) reveal that knowledge recombination novelty positively influences breakthrough innovations (β = 0.0959, β = 0.8355, β = 1.0193; all ps < 0.01). Furthermore, the results of Column (3), Column (5), and Column (7) showed that when KRN was added to the baseline model of digital technology depth, the regression coefficients for each explanatory variable remained statistically significant but exhibited a slight reduction compared to those in Column (1), Column (3), and Column (5) in Table 2 (β = 0.0115, p < 0.01; β = 0.0127, p < 0.05; β = 0.0463, p < 0.01). These findings indicate that knowledge recombination novelty serves as a partial mediator in the relationship between digital technology depth and breakthrough innovations, thereby supporting H2a.

Similarly, the results of Column (2) indicate that digital technology breadth positively affects knowledge recombination novelty (β = 0.0194, p < 0.1). The results from Column (4), Column (6), and Column (8) further confirmed that knowledge recombination novelty positively impacted breakthrough innovations (β = 0.0967, β = 0.8370, β = 1.0256, all ps < 0.01). The results of Column (4), Column (6), and Column (8) showed that the inclusion of KRN into the baseline model of digital technology breadth resulted in statistically significant regression coefficients for each explanatory variable; however, these coefficients were slightly lower than those observed in Column (2), Column (4), and Column (6), as presented in Table 2 (β = 0.0591, β = 0.0574, β = 0.1963, all ps < 0.05). Thus, H2b is supported. These results indicate that firms’ knowledge recombination novelty acts as a mediator in the relationship between digital technology adoption and breakthrough innovations.

Finally, we assessed the conditional indirect effect of the knowledge base in Table 4. The finding from Column (1) indicates that the interaction between digital technology depth and knowledge base is negatively correlated with knowledge recombination novelty (β = −0.0269, p < 0.01). In addition to the insignificant result from Column (3), where the coefficient of knowledge recombination novelty is statistically insignificant (β = −0.0249, p > 0.1), the findings from Column (5) and Column (7) demonstrated that the knowledge base suppressed the mediating effect of the knowledge recombination novelty between digital technology depth and breakthrough innovations (β = −0.1117, p < 0.01; β = −0.0887, p < 0.1), thus providing support for H3a₁ and rejecting H3a₂.

Similarly, Column (2) shows that the interaction between digital technology breadth and knowledge base is negatively correlated with knowledge recombination novelty (β = −0.4510, p < 0.05). In addition to the insignificant result from Column (4), where the coefficients of digital technology breadth and knowledge recombination novelty are statistically insignificant (β = −0.0063, β = −0.0193, both ps > 0.1), the findings from Column (6) and Column (8) demonstrated that knowledge base suppressed the mediating effect of knowledge recombination novelty between digital technology breadth and breakthrough innovations (β = −1.0605, p < 0.01; β = −1.6359, p < 0.05), thus supporting H3b₂ and disconfirming H3b₁.

We created the graphs shown in Figure 2 to more intuitively demonstrate the interaction effects. These results indicate that the positive impacts of digital technology depth and digital technology breadth on firms’ breakthrough innovations through knowledge recombination novelty are diminished when firms’ knowledge base is at a higher level.

4.2. Additional Analysis

In digitalization, enterprises have a substantial demand for knowledge and information [101,102]. Digital infrastructure, as a crucial external environmental variable in the digitalization process of enterprises, can reduce the costs associated with information search and provide diverse channels for knowledge acquisition [103]. Furthermore, digital infrastructure can enhance the combination and application of knowledge [101]. For example, it can provide opportunities for highly interactive dialogues and experience sharing among different enterprises [104], thereby promoting the exchange of tacit knowledge [105].

Although the significance of digital infrastructure in the digital transformation of enterprises has become increasingly prominent, its impact on enterprise digital innovation remains unclear in the existing literature. On the one hand, firms can leverage digital infrastructure to generate a demand-side pull effect and enhance innovation performance through the use of digital technologies [106]. On the other hand, in regions with a high level of digital infrastructure construction (DIC), firms may tend to rely on the existing environment, adopting a “free riding” approach that diminishes their investment in digital technologies [107]. In other words, a higher degree of DIC may weaken the positive effect of enterprise digitalization on innovation performance. Consequently, we posed an intriguing follow-up question: Is there a difference in the relationship between digital technology adoption and breakthrough innovations based on the level of DIC in which an enterprise is situated?

To address this issue, we conducted additional analyses by segmenting the entire sample into three subsamples based on the DIC levels of the cities where the enterprises are located: enterprises situated in cities with low levels of DIC (low-DICs), medium-DICs, and high-DICs. The level of DIC was assessed using the DIC index. To minimize subjectivity in determining indicator weights and to reduce estimation deviations due to information overlap among statistical indicators, we employed the entropy method to calculate the DIC index, utilizing six specific indicators [108]. For detailed information regarding the names, measurements, and characteristics of these indicators, see Appendix A,

Table A1. Index system for digital infrastructure construction (DIC).

Index	Indicator	Measure
DIC Index	Optical cable density (+)	The ratio of the length of long-distance optical cable lines to the area of administrative regions
	The number of Internet broadband access ports per capita (+)	The ratio of Internet broadband access ports to total population
	The number of relevant employees (+)	The proportion of employees in urban units in information transmission, computer services, and software industries
	The revenue from telecommunications services (+)	The ratio of total telecommunications revenue to total population
	The penetration of mobile phones (+)	The ratio of the number of mobile phone users to the total population
	The penetration of the Internet (+)	The ratio of the number of Internet broadband access users to the total population

Note. The “+” in parentheses denotes the indicator’s positive and negative orientation within the standardized treatment context.

.

First, we tested the influence of the level of DIC on the relationship between digital technology depth and firms’ breakthrough innovations in

Table 5. Additional analysis (DTD): Regression estimates for enterprises located in low-DICs, medium-DICs, and high-DICs.

Variables	Low-DICs			Medium-DICs			High-DICs
	Citation	Ci90	Invenum	Citation	Ci90	Invenum	Citation	Ci90	Invenum
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Size	0.0321	−0.1640 *	−0.4772 **	−0.0063	−0.1373 *	−0.2509	−0.0627	−0.1060 ***	−0.5038 ***
	(0.4662)	(−1.8306)	(−2.3357)	(−0.1138)	(−1.9249)	(−1.2666)	(−1.1885)	(−3.4143)	(−4.5673)
ROA	−0.0822	0.2404 *	0.4081	−0.0523	0.1360	1.3093 ***	−0.0583	0.2413 ***	0.1945
	(−1.1101)	(1.9547)	(0.8483)	(−0.8492)	(0.9426)	(3.1500)	(−0.9949)	(2.9297)	(0.3741)
Lev	0.0589	0.1711 *	−0.2628	0.0229	0.0201	0.1287 ***	0.0838 *	0.1270	0.0592
	(1.2506)	(1.6886)	(−1.0346)	(1.2922)	(1.0258)	(2.7469)	(1.6793)	(1.4872)	(0.2780)
Growth	0.0076	0.0361	0.0679	0.0158 *	−0.0151	0.0161	0.0092	−0.0039	0.0094
	(0.9680)	(1.1687)	(1.3100)	(1.9157)	(−0.6571)	(0.4274)	(0.9190)	(−0.2027)	(0.1939)
Cashflow	0.1420	0.5567	0.8914	0.5819 **	0.4232	3.2357 ***	0.5000 *	1.1178 *	2.6466 *
	(1.0853)	(1.4465)	(0.7278)	(2.5296)	(1.0828)	(3.6510)	(1.8769)	(1.9506)	(1.6922)
RD	−0.0002	−0.0001	−0.0012	−0.0022	0.0017	0.0402	−0.0003	−0.0003	0.0110 ***
	(−0.5344)	(−0.1892)	(−0.6919)	(−0.9143)	(0.3741)	(1.0162)	(−1.6335)	(−0.5409)	(2.9184)
DTD	0.0177 ***	0.0024	0.0225	0.0024	0.0253 **	0.0697 **	0.0132 ***	0.0178 ***	0.0485 **
	(3.7286)	(0.5093)	(1.0056)	(0.7086)	(2.3383)	(2.4752)	(2.7226)	(4.8610)	(2.3028)
Constant	0.2894 ***	0.2208 ***	0.4591 ***	0.2581 ***	0.2403 ***	0.2518 ***	0.2221 ***	0.2029 ***	0.2913 ***
	(13.5007)	(4.0882)	(4.6488)	(10.6993)	(4.8421)	(2.8498)	(8.2915)	(4.8506)	(3.5587)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	1705	1705	1705	1749	1749	1749	1705	1705	1705
R2	0.0524	0.0296	0.0991	0.0372	0.0299	0.1737	0.0950	0.0477	0.2685
Largest VIF	1.1340	1.1340	1.1340	1.2806	1.2806	1.2806	1.1012	1.1012	1.1012

Notes. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

. For enterprises located in high-DIC regions, the results from Columns (7) to (9) demonstrate that the effect of digital technology depth on breakthrough innovations is positive and statistically significant (β = 0.0132, p < 0.01; β = 0.0178, p < 0.01; β = 0.0485, p < 0.05). However, for enterprises situated in medium-DIC and low-DIC regions, the results from Columns (4) to (6) and Columns (1) to (3) indicate that the significance of the positive impact of digital technology depth on breakthrough innovations is gradually decreasing.

Second, we examined the impact of the level of DIC on the relationship between digital technology breadth and firms’ breakthrough innovations in

Table 6. Additional analysis (DTB): Regression estimates for enterprises located in low-DICs, medium-DICs, and high-DICs.

Variables	Low-DICs			Medium-DICs			High-DICs
	Citation	Ci90	Invenum	Citation	Ci90	Invenum	Citation	Ci90	Invenum
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Size	0.0318	−0.1834 **	−0.4846 **	−0.0121	−0.1478 **	−0.2825	−0.0675	−0.1102 ***	−0.5192 ***
	(0.4438)	(−2.1405)	(−2.1912)	(−0.2157)	(−1.9913)	(−1.3995)	(−1.2177)	(−3.5998)	(−4.7216)
ROA	−0.0986	0.2508 **	0.3920	−0.0473	0.1330	1.3041 ***	−0.0500	0.2481 ***	0.2210
	(−1.2534)	(2.0335)	(0.8412)	(−0.7661)	(0.9176)	(3.1271)	(−0.8888)	(2.9440)	(0.4208)
Lev	0.0488	0.1705*	−0.2753	0.0247	0.0192	0.1270 ***	0.1010 *	0.1396	0.1126
	(0.9216)	(1.6926)	(−1.0727)	(1.3017)	(0.9682)	(2.6343)	(1.9121)	(1.6248)	(0.5203)
Growth	0.0085	0.0348	0.0685	0.0147 *	−0.0167	0.0111	0.0082	−0.0042	0.0066
	(1.0661)	(1.1212)	(1.3430)	(1.8390)	(−0.7351)	(0.3008)	(0.8779)	(−0.2199)	(0.1412)
Cashflow	0.1581	0.4910	0.8871	0.5744 **	0.4806	3.3863 ***	0.3960	1.0062 *	2.2911
	(1.2119)	(1.3910)	(0.7072)	(2.5061)	(1.2201)	(3.8365)	(1.4819)	(1.7579)	(1.4842)
RD	−0.0002	−0.0000	−0.0012	−0.0020	0.0016	0.0400	−0.0003 **	−0.0004	0.0108 ***
	(−0.5876)	(−0.0256)	(−0.7119)	(−0.8227)	(0.3457)	(1.0165)	(−2.1854)	(−0.6927)	(2.9323)
DTB	0.0148	0.0969	0.0534	0.0266	0.0362	0.1131	0.1079 **	0.0935 *	0.3483 **
	(0.4912)	(1.4180)	(0.2810)	(0.8679)	(0.8102)	(0.8614)	(2.3734)	(1.7774)	(2.2816)
Constant	0.2915 ***	0.2188 ***	0.4609 ***	0.2569 ***	0.2392 ***	0.2481 ***	0.2208 ***	0.2024 ***	0.2878 ***
	(13.4391)	(4.0702)	(4.6587)	(10.6039)	(4.7853)	(2.7640)	(8.4085)	(4.8345)	(3.5098)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	1705	1705	1705	1749	1749	1749	1705	1705	1705
R2	0.0329	0.0322	0.0982	0.0387	0.0238	0.1648	0.1087	0.0451	0.2706
Largest VIF	1.1365	1.1365	1.1365	1.2806	1.2806	1.2806	1.1158	1.1158	1.1158

Notes. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

. For enterprises located in High-DICs, the results from Columns (7) to (9) demonstrated that the effect of digital technology breadth on breakthrough innovations was positive and statistically significant (β = 0.1079, p < 0.05; β = 0.0935, p < 0.1; β = 0.3483, p < 0.05). Furthermore, compared to the corresponding models in Table 5, the regression coefficient for digital technology breadth was somewhat larger than that for digital technology depth. In contrast, for enterprises located in medium-DICs and low-DICs, the results from Columns (4) to (6) and Columns (1) to (3) demonstrate that the effect of digital technology breadth on breakthrough innovations is positive but statistically insignificant.

These findings suggest that the positive impacts of digital technology adoption on breakthrough innovations are stronger for enterprises located in high-DIC regions compared to those in medium-DIC and low-DIC regions. One possible explanation for this observation is that a high level of DIC in a region indicates the establishment of a mature operational system for the creation, use, and sharing of data, which, in turn, transforms data into an effective information resource. Through digital technologies, enterprises can efficiently acquire the necessary information and knowledge, facilitating breakthrough innovations. Overall, our research demonstrates that the relationship between digital technology adoption and breakthrough innovations is more pronounced for enterprises located in high-DIC regions.

4.3. Endogeneity

In the previous main models, we employed various indicators of breakthrough innovation and implemented a one-year time lag between digital technology adoption and breakthrough innovations to mitigate estimation bias. However, this study may still be subject to endogeneity issues arising from omitted variables. We treated digital technology adoption as endogenous and utilized the average digital technology depth (Avg_DTD) and digital technology breadth (Avg_DTB) of firms outside the industry segment to which the observed firms belong as instrumental variables to correct for potential biases. Specifically, industry competition tends to stimulate enterprises participating within the industry to adopt emerging technologies [109]. When other enterprises within the industry adopt digital technologies, the competitive pressure and uncertainty generated by changes in the external environment will compel observing enterprises to adjust their investment strategies for digital technologies [110]. However, while the investment decisions of other enterprises may influence the investment choices of the observed enterprises, they do not have a direct impact on the breakthrough innovation performance of those observed enterprises. Therefore, Avg_DTD and Avg_DTB meet the requirements of instrumental variables.

To estimate the influence of endogeneity on digital technology adoption, we employed a two-stage least squares (2SLS) regression model. As presented in

Table 7. Endogeneity test: 2SLS estimation results.

Variables	First-Stage		Second-Stage
	DTD	DTB	Citation		Ci90		Invenum
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Size	0.0049	0.1068 ***	−0.0265	−0.0419 **	−0.1190 ***	−0.1579 ***	−0.4127 ***	−0.4194 ***
	(0.0509)	(4.3738)	(−1.4482)	(−2.1386)	(−4.2322)	(−5.1800)	(−5.3531)	(−5.1337)
ROA	−0.6924 ***	−0.1552 ***	−0.0392	−0.0445 *	0.2934 ***	0.2666 ***	0.6886 **	0.5987 **
	(−3.1543)	(−2.8132)	(−1.4994)	(−1.6921)	(4.1598)	(4.1330)	(2.5117)	(2.2626)
Lev	−0.2784 **	−0.0958 ***	0.0429 ***	0.0448 ***	0.0963 ***	0.0950 ***	0.0651	0.0249
	(−2.4435)	(−2.9690)	(2.8295)	(2.8439)	(3.0848)	(3.0682)	(1.1972)	(0.4116)
Growth	0.0552	0.0265 *	0.0149 **	0.0134 **	0.0066	0.0044	0.0379	0.0484
	(1.3201)	(1.7817)	(2.3370)	(2.0185)	(0.2820)	(0.1834)	(0.9352)	(1.1610)
Cashflow	0.5761	0.6124 ***	0.3746 ***	0.3225 ***	0.5951 **	0.4863 *	2.1776 ***	2.3052 ***
	(0.7164)	(3.2280)	(3.7253)	(3.1468)	(2.3261)	(1.9347)	(3.5555)	(3.6757)
Rd	−0.0023 **	−0.0002	−0.0001	−0.0002 *	0.0004	0.0001	0.0053	0.0047
	(−2.2968)	(−0.3189)	(−1.1532)	(−1.7691)	(1.0681)	(0.2625)	(1.5172)	(1.3313)
DTD			0.0390 ***		0.1234 ***		0.1809 **
			(2.6227)		(2.7750)		(2.0099)
IV: Avg_DTD	0.4607 ***
	(4.0670)
DTB				0.1296 ***		0.3238 **		0.0333
				(2.6912)		(2.3937)		(0.1169)
IV: Avg_DTB		0.5264 ***
		(7.0717)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5049	5049	5049	5049	5049	5049	5049	5049
R2			−0.0111	0.0374	−0.1502	−0.0015	0.1258	0.1641
Largest VIF	1.1180	1.1170	1.1110	1.1110	1.1110	1.1110	1.1110	1.1110
LM. p-value	0.0001	0.0000
Cragg–Donald Wald F	16.541	50.009

Notes. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

, the first-stage regression results (Column (1) and Column (2)) indicated that the coefficients of the instrumental variables were both positive and statistically significant (β = 0.4607, β = 0.5264; both ps < 0.01). The p-values from the Anderson LM test were 0.0001 and 0.0000, which clearly denies the concerns regarding unrecognized instrumental variables. The Cragg–Donald Wald F statistics from the weak identification test were 16.541 and 50.009, both exceeding the critical value of the Stock–Yogo Weak ID test (16.38), which strongly rejects the null hypotheses of “weak instrumental variables”. These results imply that our instrumental variables are valid.

In the second-stage regression, the findings from Column (3), Column (5), and Column (7) indicated a positive correlation between digital technology depth and breakthrough innovations (β = 0.0390, p < 0.01; β = 0.1234, p < 0.01; β = 0.1809, p < 0.05). The results from Column (4), Column (6), and Column (8) further demonstrate that with the exception of Column (8) (β = 0.0333, p > 0.1), digital technology breadth was positively correlated with breakthrough innovations (β = 0.1296, p < 0.01; β = 0.3238, p < 0.05). Therefore, H1a and H1b receive additional support.

4.4. Robustness Check

4.4.1. Alternative Explanatory Variables

We employed alternative measures for the independent variables. In the previous main models, the independent variables (DTD and DTB) were measured by the word frequency of the four types of digital technology (DT): artificial intelligence, blockchain, cloud computing, and big data. Considering that Li et al. [111], Urbinati et al. [112], and Zhou et al. [15] selected keywords related to digital technology from the two dimensions of “underlying DT” and “DT practice”, in order to eliminate possible measurement bias, we added keywords related to “DT practice” to the original word frequency statistics table, such as industrial internet, intelligent transportation systems, and digital finance. As illustrated in

Table 8. Robustness check: Regression estimates for alternating explanatory variables.

Variables	Ci90						KRN
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Size	−0.1235 ***	−0.0826 **	−0.0547 **	−0.1305 ***	−0.0889 ***	−0.0511 ***	−0.1226 ***	−0.0650 ***	−0.1254 ***	−0.0608 ***
	(−3.9506)	(−2.1708)	(−2.1664)	(−4.0770)	(−3.0932)	(−2.6496)	(−5.9812)	(−5.4438)	(−11.8772)	(−9.2951)
ROA	0.2163 ***	0.2747 **	0.0560	0.2146 ***	0.2701 **	0.0521	0.2561 ***	0.0730 ***	0.2594 ***	0.0707 ***
	(3.3301)	(2.2812)	(1.0323)	(3.2882)	(2.3876)	(0.7505)	(6.0304)	(3.0713)	(6.1720)	(2.9797)
Lev	0.0636 **	−0.0308	0.0210	0.0641 **	−0.0382	0.0227	0.1010 ***	0.0364	0.1020 ***	0.0366 ***
	(2.0430)	(−0.4802)	(1.1525)	(2.0341)	(−0.6252)	(0.7341)	(3.8421)	(1.3290)	(4.4874)	(3.4615)
Growth	0.0071	0.0345	0.0064	0.0069	0.0374	0.0053	0.0232	−0.0033	0.0221	−0.0036
	(0.4089)	(0.7540)	(0.3836)	(0.3959)	(0.9510)	(0.4311)	(1.2268)	(−0.8947)	(1.5090)	(−0.8497)
Cashflow	0.6667 ***	0.4263	0.2691	0.6499 **	0.4088 **	0.2563	0.4195 ***	0.2399 ***	0.4082 ***	0.2414 ***
	(2.6378)	(1.4916)	(1.1642)	(2.5864)	(1.9613)	(1.5501)	(3.7187)	(3.2437)	(5.2657)	(4.2764)
RD	0.0000	0.3185 **	−0.0005 **	−0.0000	0.3648 **	−0.0006	0.2104 ***	0.0005 ***	0.2219 ***	0.0005
	(0.0289)	(2.3025)	(−2.3437)	(−0.1413)	(2.2360)	(−0.6541)	(4.2662)	(5.3398)	(3.6516)	(1.5772)
DTD	0.0210 ***	0.0182 ***	0.0266 ***				0.0033 *	0.0025 **
	(3.6914)	(3.3159)	(3.7637)				(1.8707)	(2.1157)
DTB				0.0374 *	0.0282 *	0.0587 ***			0.0127 **	0.0108 *
				(1.6691)	(1.7423)	(3.0979)			(2.1036)	(1.6698)
KRN		0.8180 ***	0.5458 ***		0.8213 ***	0.5388 ***
		(13.3191)	(7.2457)		(20.9399)	(12.5966)
KB			4.4236 ***			4.8195 ***		3.1656 ***		3.3343 ***
			(7.2007)			(15.0736)		(10.2642)		(34.0648)
DTD ✕ KB			−0.1898 ***					−0.0477 ***
			(−4.6628)					(−3.1377)
DTB ✕ KB						−0.8997 ***				−0.3621 ***
						(−4.0768)				(−4.8088)
Constant	0.2272 ***	0.1108 ***	0.0813 ***	0.2262 ***	0.1105 ***	0.0757 ***	0.1166 ***	0.0852 ***	0.1157 ***	0.0828 ***
	(8.0919)	(4.0121)	(2.9334)	(7.9803)	(4.0530)	(2.9199)	(11.8920)	(7.8550)	(11.5421)	(9.4216)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5159	5159	5159	5159	5159	5159	5159	5159	5159	5159
R2	0.0294	0.1194	0.1577	0.0257	0.1163	0.1564	0.1394	0.3061	0.1394	0.3090
Largest VIF	1.1111	1.1830	2.0049	1.1114	1.2471	2.5084	1.1458	1.9891	1.1963	2.4796

Notes. We used Ci90 to represent breakthrough innovation to save space within the table. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

, the results of this robustness test indicate that (1) digital technology adoption positively influences breakthrough innovations; (2) knowledge recombination novelty serves as a mediator in the relationship between digital technology adoption and breakthroughs; and (3) knowledge base moderates the link between digital technology adoption, knowledge recombination novelty, and breakthrough innovations. These findings further validate the conclusions derived from the primary results.

4.4.2. Bootstrap

We employed the bootstrap method to verify the mediating role of knowledge recombination novelty in the link between digital technology adoption and breakthrough innovations. The results of this robustness test, presented in

Table 9. Robustness check: bootstrap results for mediating effects of knowledge recombination novelty.

Independent Variables	Mediator Variable	Direct Effect				Indirect Effect
		Effect	SE	95% CI		Effect	SE	95% CI
				LL	UL			LL	UL
DTD	KRN	0.0127	0.0049	0.0005	0.0212	0.0031	0.0013	0.0014	0.0064
DTB	KRN	0.0574	0.0259	0.0074	0.1067	0.0162	0.0075	0.0014	0.0316

Notes. The bootstrap resampling method was conducted with 1000 iterations, where SE denotes the standard error, and CI represents the confidence interval. LL refers to the lower limit, while UL indicates the upper limit.

, indicate that digital technology depth exerts a direct influence on breakthrough innovations (β = 0.0127, 95% confidence interval (CI) = [0.0005, 0.0212]) and an indirect influence via knowledge recombination novelty (β = 0.0031, 95% CI = [0.0014, 0.0064]). Similarly, digital technology breadth demonstrates a direct impact on breakthrough innovations (β = 0.0574, 95% CI = [0.0074, 0.1067]) and an indirect impact via knowledge recombination novelty (β = 0.0162, 95% CI = [0.0014, 0.0316]). Therefore, H2a and H2b are further supported.

We also employed the bootstrap method to verify the conditional indirect effect of the knowledge base and examine the mediating role of knowledge recombination novelty between digital technology adoption and breakthrough innovations at different levels of the knowledge base. The results of this robustness test, presented in

Table 10. Robustness check: bootstrap results for conditional indirect effects of knowledge base.

Independent Variables	Mediator Variable	Moderator Variable	Condition of Moderator Variable	Conditional Indirect Effect
				Effect	SE	95% CI	Effect
						LL	UL
DTD	KRN	KB	Mean − SD	0.0015	0.0008	0.0005	0.0039
			Mean	0.0008	0.0007	0.0001	0.0029
			Mean + SD	0.0001	0.0006	−0.0008	0.0014
DTB	KRN	KB	Mean − SD	0.0150	0.0056	0.0014	0.0252
			Mean	0.0031	0.0037	−0.0033	0.0111
			Mean + SD	−0.0088	0.0047	−0.0183	0.0021

Notes. The bootstrap resampling method was conducted with 50 iterations, where SE denotes the standard error, and CI represents the confidence interval. LL refers to the lower limit, while UL indicates the upper limit. Mean and SD denote the mean and standard deviation of KB.

, indicate that when the level of knowledge base is low, the indirect impact of digital technology depth on firms’ breakthrough innovations through knowledge recombination novelty is statistically significant (β = 0.0015, 95% CI = [0.0005, 0.0039]). In contrast, when the level of knowledge base is high, this indirect impact is not statistically significant (β = 0.0001, 95% CI = [−0.0008, 0.0014]).

Similarly, for firms with a low knowledge base, the indirect influence of digital technology breadth on breakthrough innovations through knowledge recombination novelty is statistically significant (β = 0.0150, 95% CI = [0.0014, 0.0252]), whereas, at a high knowledge base level, the indirect influence of digital technology breadth on breakthrough innovations is not statistically significant (β = −0.0088, 95% CI = [−0.0183, 0.0021]). Overall, as the level of knowledge base increases, the indirect effects and significance of the relationship between digital technology adoption and firms’ breakthrough innovations through knowledge recombination novelty gradually diminish. Thus, H3a₁ and H3b₂ are further supported.

4.4.3. Alternative Model Specification

Finally, we re-evaluated the results by modifying the regression model. Due to resource constraints and competitive pressures, many enterprises in China refrain from applying for patents, leading to zero values for the dependent variables in numerous observations within the sample. According to Amemiya [113], we utilized the Tobit model to estimate the effects of digital technology adoption on breakthrough innovations, thereby addressing the regression bias associated with this truncated data. As shown in

Table 11. Robustness check: regression estimates for alternating model specification.

Variables	Ci90						KRN
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Size	−0.3300 ***	−0.1932 ***	−0.1814 ***	−0.3477 ***	−0.2019 ***	−0.1799 ***	−0.1327 ***	−0.1149 ***	−0.1363 ***	−0.1138 ***
	(−6.1949)	(−4.2266)	(−3.7941)	(−6.4710)	(−4.4005)	(−3.8369)	(−6.8609)	(−6.8532)	(−6.9701)	(−7.4867)
ROA	2.3312 ***	1.6025 ***	1.3006 ***	2.2903 ***	1.5764 ***	1.2760 ***	0.9235 ***	0.6243 ***	0.9137 ***	0.6443 ***
	(4.6084)	(3.2525)	(2.7780)	(4.5463)	(3.2104)	(2.7404)	(6.9811)	(5.9614)	(6.9225)	(5.9247)
Lev	0.8584 ***	0.4867 **	0.4556 **	0.8600 ***	0.4846 **	0.4678 **	0.1455 ***	0.0971 **	0.1461 ***	0.0901 ***
	(3.6943)	(2.2170)	(2.1203)	(3.6926)	(2.2018)	(2.1734)	(2.8215)	(2.2947)	(2.8108)	(3.0130)
Growth	−0.0077	0.0119	0.0113	−0.0095	0.0113	0.0091	−0.0182	−0.0130	−0.0185	−0.0132
	(−0.1588)	(0.2704)	(0.2567)	(−0.1949)	(0.2568)	(0.2041)	(−1.4888)	(−1.2599)	(−1.5013)	(−1.1815)
Cashflow	1.8400 ***	0.8241	0.7314	1.7674 ***	0.7851	0.6683	0.8845 ***	0.6890 ***	0.8758 ***	0.6807 ***
	(3.3022)	(1.5522)	(1.4459)	(3.1793)	(1.4786)	(1.3209)	(5.7364)	(5.1364)	(5.7025)	(4.9486)
RD	−0.0125 **	−0.0111 **	−0.0092 **	−0.0122 **	−0.0109 **	−0.0090 **	−0.0022	−0.0011	−0.0021	−0.0009
	(−2.4099)	(−2.4874)	(−2.2006)	(−2.3236)	(−2.4235)	(−2.1454)	(−1.4746)	(−0.9861)	(−1.4315)	(−0.1468)
DTD	0.0311 ***	0.0184 **	0.0331 ***				0.0098 ***	0.0112 ***
	(3.8839)	(2.3651)	(3.8745)				(3.2204)	(2.6643)
DTB				0.1551 ***	0.0802	0.1695 **			0.0431 **	0.0434 *
				(2.6144)	(1.4400)	(2.0309)			(2.3162)	(1.7104)
KRN		2.2576 ***	1.5136 ***		2.2642 ***	1.5183 ***
		(18.9923)	(10.8096)		(19.0590)	(10.7750)
KB			11.6508 ***			11.9460 ***		6.0487 ***		6.2484 ***
			(10.5197)			(10.2347)		(13.5119)		(28.0492)
DTD ✕ KB			−0.2066 ***					−0.0547 ***
			(−5.6710)					(−3.1556)
DTB ✕ KB						−1.9718 **				−0.6100 **
						(−2.4952)				(−2.2474)
Constant	−0.6945	−1.1882 **	−1.3307 ***	−0.6626	−1.1728 **	−1.3525 ***	0.1882 **	0.0139	0.1953 **	−0.1076
	(−1.3198)	(−2.1473)	(−2.7432)	(−1.2664)	(−2.1223)	(−2.7919)	(2.3455)	(0.1648)	(2.4427)	(−0.9665)
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5159	5159	5159	5159	5159	5159	5159	5159	5159	5159
F. p-value	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
Largest VIF	1.1110	1.1278	2.1422	1.1110	1.1278	2.1965	1.1110	2.1340	1.1110	2.1780

Notes. We used Ci90 to represent breakthrough innovation to save space within the table. Robust t-statistics are reported in parentheses. * p < 0.10; ** p < 0.05; *** p < 0.01.

, the results of this analysis are consistent with our earlier findings.

5. Discussion

This study presents three main findings. First, we found that the two distinct dimensions of digital technology adoption—depth and breadth—positively influenced breakthrough innovations. This is in line with KBV, which suggests that knowledge serves as a core resource for firms and significantly influences breakthrough innovation [16]. By adopting digital technologies, firms can transform data into knowledge, increasing the likelihood of breakthrough innovations. A substantial body of current research has concentrated on the effect of various types of digital technologies on firms’ innovation performance (e.g., Usai et al. [14]), yet few empirical studies have examined the effects of digital technology adoption through the lens of specific strategies related to digital technology depth and digital technology breadth. This study builds upon previous research (e.g., Blichfeldt and Faullant [10]; Yang et al. [37]) by expanding the investigation of digital technology adoption into the realm of breakthrough innovation. Thus, this study offers a more vivid investigation of the antecedents of breakthrough innovation.

Furthermore, this study responds to the call for a holistic exploration of the relationship between digital technology and breakthrough innovation [35], advocating for the consideration of a digital technology portfolio rather than focusing solely on individual technologies. In this way, our study complements prior research on strategic methodologies intended to enhance breakthrough innovation performance.

Second, we demonstrated that knowledge recombination novelty mediates the relationship between digital technology adoption and breakthrough innovations. This result suggests that digital technology adoption is a key enabler of knowledge production, particularly by enhancing knowledge recombination novelty, which, in turn, leads to breakthrough innovations in the form of patent inventions. While scholars have reached a consensus regarding the relationship between knowledge recombination novelty and breakthrough innovation (e.g., Arts and Veugelers [21]; Zhong et al. [64]), there is still a notable lack of empirical studies that examine this relationship within the context of digital technology. Our study extends beyond the conventional practice of treating knowledge recombination novelty as a dependent variable, as seen in previous innovation studies (e.g., Luo et al. [114]). We identify knowledge recombination novelty as a mediating factor, as the depth and breadth of digital technology furnish firms with diverse knowledge resources that are crucial for cultivating their knowledge recombination novelty. This, in turn, enhances their capacity for breakthrough innovations (e.g., Ferreras-Méndez et al. [115]; Forman and Van Zeebroeck [22]). Therefore, this study addresses the existing research gap regarding how the adoption of digital technology contributes to breakthrough innovations, specifically in the form of patent inventions, by empirically testing knowledge recombination novelty as an underlying mechanism.

Third, we found that the knowledge base—a unique strategic resource of enterprises—negatively moderates the mediating effect of knowledge recombination novelty between digital technology adoption and breakthrough innovations. Specifically, enterprises with a lower knowledge base are more likely to achieve digital technology-driven breakthrough innovations through knowledge recombination. Conversely, a high knowledge base appears to hinder such innovations. To the best of our knowledge, in the field of knowledge management, recent studies on digital technology have predominantly concentrated on how firms utilize digital technology to acquire external knowledge for innovation (e.g., Qu et al. [7]; Guo et al. [9]); however, there is limited understanding of how internal knowledge acts as a contingency factor that enhances the adoption of digital technology in fostering breakthrough innovations. Our result affirms the complementary relationship between digital technology and the knowledge base, consistent with Ko and Liu [73], who argued that a firm’s internal knowledge could serve as a contingency factor to complement its information technology competency in promoting innovation. Therefore, this finding suggests that in the future, firms seeking to achieve digital technology-driven breakthrough innovations through knowledge recombination should adopt digital technology investment strategies tailored to their level of knowledge base.

6. Conclusions and Implications

6.1. Conclusions

This study investigates how the depth and breadth of digital technology adoption influence firms’ breakthrough innovations from the perspective of KBV. It utilizes panel data collected from 469 publicly listed biomedical companies in China. The findings indicate that both dimensions of digital technology adoption are positively associated with breakthrough innovations, with these relationships mediated by the internal mechanism of knowledge recombination novelty. Furthermore, the interactions between both dimensions of digital technology adoption and the knowledge base are negatively correlated with knowledge recombination novelty. Additionally, further analyses demonstrate that the impacts of digital technology adoption on breakthrough innovations vary among firms at different levels of DIC. Specifically, the positive effects of digital technology adoption on breakthrough innovations are more pronounced for enterprises situated in high-DIC environments compared to those in low-DIC and medium-DIC contexts.

6.2. Managerial Implications

This study presents three actionable recommendations for business managers and policymakers. First, in the digital era, managers aiming to obtain breakthrough innovations to sustain their firms’ competitive advantage may do so by effectively managing their digital technology investment strategies. Regarding digital technology depth, firms should aim to enhance their proficiency in utilizing various types of digital technologies. This will enable them to process, analyze, and exploit knowledge derived from data to uncover hidden market opportunities and implement ideas from emerging technologies and market areas. In terms of digital technology breadth, firms should expand their existing digital technology portfolio to the greatest extent possible. This expansion will facilitate access to a diverse array of knowledge sources and enhance knowledge accumulation, thereby enabling firms to acquire exclusive knowledge and achieve breakthrough innovations.

Second, enterprises should prioritize knowledge recombination within their organizational practices, enabling them to integrate knowledge resources from different sources and select suitable transformation paths based on their needs. This study indicates that managers must acknowledge that achieving breakthrough innovations necessitates not only a deep knowledge base but also strong recombinant creativity, which involves the challenging task of synthesizing knowledge from various sources. For instance, to address the challenges encountered by team members during the knowledge recombination process, managers should, on the one hand, guide team members in actively decoding and reconstructing heterogeneous knowledge. On the other hand, they should enhance the existing knowledge management system to facilitate knowledge sharing and discussion among team members.

Third, firms should strategically implement their investment strategies in digital technology in alignment with their knowledge base level to achieve effective complementarity between their knowledge base and digital technology. For example, to enhance the sophistication and complexity of their knowledge, firms should focus on investing in a specific type of digital technology. Conversely, to broaden the diversity and scope of their knowledge, firms should make investments across multiple areas of digital technology. Furthermore, governments should fully leverage digital technologies by creating digital ecosystems that enable firms to achieve breakthrough innovations. For instance, to encourage firms with low knowledge base levels to attain digital technology-driven breakthrough innovations through knowledge recombination, governments should prioritize the enhancement of digital infrastructure in the regions where these firms operate.

6.3. Limitations and Future Directions

This study has three limitations that offer directions for future research. First, while this research examined the roles of knowledge recombination novelty and knowledge base in the relationship between digital technology adoption and breakthrough innovations, it is recommended that future researchers explore additional factors that may influence this relationship, such as change management (e.g., Chatzinikolaou and Vlados [116]), corporate social responsibility (CSR) (e.g., Wang et al. [117]), and environmental dynamism (e.g., Zhong et al. [64]). Investigating these factors could further expand the discourse on this topic. Second, our findings were exclusively based on panel data collected in China. Future research could broaden this investigation to include diverse contexts, encompassing other countries and industries not examined in this study. Third, this research was conducted at the firm level and did not address complex cross-level mechanisms (e.g., individual and team levels). Future research could consider developing a multilevel model based on the existing literature to enrich and refine the theoretical framework presented in this paper.