Exploring Influential Factors with Structural Equation Modeling–Artificial Neural Network to Involve Medicine Users in Home Medicine Waste Management and Preventing Pharmacopollution

Abstract

The appropriate management of home medical waste is of paramount importance due to the adverse consequences that arise from improper handling. Incorrect disposal practices can lead to pharmacopollution, which poses significant risks to environmental integrity and human well-being. Involving medicine users in waste management empowers them to take responsibility for their waste and make informed decisions to safeguard the environment and public health. The objective of this research was to contribute to the prevention of pharmacopollution by identifying influential factors that promote responsible disposal practices among medicine users. Factors such as attitude, marketing campaigns, collection points, safe handling, medical prescription, package contents, and public policies and laws were examined. To analyze the complex relationships and interactions among these factors, a dual-staged approach was employed, utilizing advanced statistical modeling techniques and deep learning artificial neural network algorithms. Data were collected from 952 respondents in Pernambuco, a state in northeastern Brazil known for high rates of pharmacopollution resulting from improper disposal of household medical waste. The results of the study indicated that the propositions related to safety in handling and medical prescription were statistically rejected in the structural equation modeling (SEM) model. However, in the artificial neural network (ANN) model, these two propositions were found to be important predictors of cooperative behavior, highlighting the ANN’s ability to capture complex, non-linear relationships between variables. The findings emphasize the significance of user cooperation and provide insights for the development of effective strategies and policies to address pharmacopollution.

1. Introduction

The generation of hazardous home waste medicine (HWM) has increased due to various factors, including the global rise in medication usage, primarily driven by the aging population and the emergence of new diseases or mental illnesses [1,2,3]. The motivations behind HWM generation include non-adherence to medical treatment, treatment abandonment, allergic reactions to prescribed medications, availability of excessive medication packaging, distribution of free samples, product obsolescence, and difficulties in remembering the storage location of medicines at home [4,5,6]. One of the main consequences of mishandled HWM is pharmacopollution [7].

Pharmacopollution, a type of pollution resulting from the presence of pharmaceutical compounds, poses a significant environmental and public health concern. It involves the release of active pharmaceutical ingredients (APIs) through direct or indirect medication pollution, as well as the excretion of endocrine-disrupting compounds (EDCs) during treatment [8,9]. These compounds, characterized by their physical and chemical properties, have the potential to adversely affect both aquatic and terrestrial environments, thereby posing risks to human, animal, and plant life [10].

The impact of pharmacopollution is particularly evident in water sources. Traces of pharmaceutical contaminants have been detected in various water bodies, including drinking water supplies, surface waters, and groundwater, exacerbating the scale of the problem [11,12,13,14]. Prolonged exposure to pharmacopollution can have serious implications. It has been linked to genetic mutations, toxic effects, the development of cancerous cells, reproductive disorders, and the emergence of antibiotic resistance, compromising the efficacy of crucial therapeutic interventions [15,16,17,18,19].

In addition to the release of HWM into the environment, pharmacopollution can occur through three primary pathways: disposal in bathroom sinks, both in solid and liquid forms [20]; improper disposal in regular garbage without prior treatment or separation; and the excrement of patients undergoing treatment [21]. The decision-making process of users regarding their involvement in waste management is complex and influenced by various factors such as economic, social, environmental, and psychological considerations [22]. Users tend to be less cooperative when their involvement is not actively requested or when they do not perceive sufficient benefits that align with their personal values [23].

Efforts to promote medicine users’ involvement and cooperation in HWM management have identified several strategies, including the donation of unused but still valid medicines, exchanging outdated medication for newly prescribed ones, resale of unused valid medicines, and proper disposal through reverse logistics [24]. However, regulatory bodies discourage the first three forms of management due to the challenges associated with ensuring purity, proper storage, and potential risks of self-medication and chemical dependency, which further exacerbate the issue of pharmacopollution and HWM [25].

The effective management of HWM largely depends on the willingness of users to cooperate, which is influenced by their level of awareness regarding the efficacy of their actions [23]. Users who are less inclined to cooperate with HWM management contribute to lower and irregular return rates of HWM, leading to a higher risk of pharmacopollution [26].

Studies in the literature often utilize the Theory of Planned Behavior (TPB) to investigate the determinants that influence user cooperative behavior [27,28]. However, it has been argued that relying solely on TPB determinants may not be sufficient, as these determinants focus on deliberate and intentional choices. Therefore, it is important to identify both intentional and unintentional creative factors that influence user involvement cooperating in HWM management [29,30].

Hence, the primary aim of this investigation was to propose a deep learning dual-staged model, combining structural equation modeling (SEM) with an artificial neural network (ANN), termed SEM-ANN, to comprehensively explore intentional and unintentional factors influencing user involvement and cooperation in the proper management of household waste medicine, thus mitigating the risk of pharmacopollution. This study highlights the utilization of SEM, a modeling technique employed to assess the validity of theoretical models that establish hypothetical causal relationships among variables, which has been relatively underexplored in the domain of household waste medicine and pharmacopollution [31].

In addition to the prior contributions of SEM, conventional surveys utilizing statistical methodologies such as multiple regression and SEM have encountered limitations in capturing solely linear relationships between variables. To address this limitation, our study adopts an integrated dual-staged approach that integrates deep learning techniques with both an SEM and ANN, as recommended by [32], offering the potential for more comprehensive and insightful findings compared to traditional ANN methods [33,34,35].

The structure of this paper encompasses several sections. The introductory section (Section 1) provides an overview of the study’s objectives. The subsequent section (Section 2) delineates the materials and methods employed for data collection and analysis. The following section (Section 3) presents the results and engages in their detailed discussion. Lastly, in the concluding section (Section 4), key insights are summarized, conclusions are drawn, and implications for theory and practice are discussed, including recommendations for future research endeavors.

2. Materials and Methods

2.1. Hypothesis Formulation

Based on an extensive review of the available literature encompassing published articles, regulations, and other relevant sources, this study formulated seven hypotheses regarding HWM and pharmacopollution.

2.1.1. The User’s Attitude

The proper management of home waste medicine (HWM) relies on the attitude of medicine users, which involves their actions based on evaluations of behaviors as favorable or unfavorable. These evaluations consider the social, environmental, and economic consequences of HWM management [36]. Studies by [37] indicate a positive association between user cooperation in returning and disposing of HWM and their awareness of the environmental impact. Recognizing the adverse consequences of improper HWM disposal motivates users to be involved and seek correct alternatives [38]. Based on these findings, the first hypothesis was formulated.

Hypothesis 1 (H1). 

Attitude positively influences the cooperative behavior of the user to properly manage the HWM in order to avoid pharmacopollution.

2.1.2. Marketing Campaigns

Marketing plays a crucial role in influencing user behavior by creating value and establishing favorable relationships [39]. It shapes user attitudes and engagement toward specific products, services, or processes [40]. Environmentally friendly marketing campaigns can inform users about necessary actions for environmental preservation and differentiate products in the market [41]. These campaigns have a positive effect on waste management by guiding users’ involvement toward proper household waste medicine (HWM) practices [42]. Thus, the second hypothesis focuses on the impact of marketing campaigns on user cooperative behavior in HWM management.

Hypothesis 2a (H2a). 

Marketing campaigns positively influence user cooperative behavior to properly manage HWM, thus avoiding pharmacopollution.

Hypothesis 2b (H2b). 

Marketing campaigns positively moderate the user’s attitude to properly manage HWM, thus avoiding pharmacopollution.

2.1.3. Collection Points

When users face difficulties in accessing appropriate collection points for the disposal of household waste medicine (HWM), their motivation to properly dispose of HWM may decrease [43]. This can occur when users encounter challenges in locating and accessing collection points for end-of-life (EOL) or end-of-use (EOU) disposal of medicines. To address this issue, strategies such as implementing home collection services and establishing widely distributed collection points have been employed in certain countries like the United Kingdom [44]. Studies have shown that the proximity of collection points to users positively influences their engagement and active participation in the management and proper disposal of household waste [45]. Based on these insights, the third hypothesis was formulated.

Hypothesis 3 (H3). 

Ease of access to collection points positively influences the cooperative behavior of the user to properly manage their HWM, thus avoiding pharmacopollution.

2.1.4. Handling Safety

Awareness of the negative consequences of improper HWM management on public health can make users hesitant and uncomfortable when it comes to handling and sending HWM to collection points [46]. Governments in the European Union and the United States of America have implemented programs aimed at collecting and disposing of EOL or EOU medicines, as well as providing information to users on safe handling practices and proper disposal methods for HWM [47]. However, when users are not aware of these recommended practices, they may feel uncertain and apprehensive about handling HWM and ensuring its proper disposal. Based on these considerations, the fourth hypothesis was formulated.

Hypothesis 4 (H4). 

Handling safety negatively influences the user’s cooperative behavior to properly manage the HWM, thus avoiding pharmacopollution.

2.1.5. Medical Prescription

The issuance of medical prescriptions that exceed users’ needs is identified as a contributing factor to the increased generation of HWM [44]. Users’ non-adherence to treatment, driven by perceived adverse effects and prescribed medication quantities, also significantly contributes to HWM generation [48]. Additionally, prescription medicines often reach their expiration date without deteriorating, but users discontinue their use when doctors modify treatments, prescribing alternative medications that effectively address their health issues [49]. Based on these findings, the fifth hypothesis was formulated.

Hypothesis 5 (H5). 

The medical prescription negatively influences the cooperative behavior of the user to adequately manage the HWM, thus avoiding pharmacopollution.

2.1.6. Package Contents

The quantity of medication purchased by users is influenced by the amount available in the packages and the incentives provided, such as discounts for bulk purchases. However, this practice is considered a major contributor to the increased stockpiling of medication within households, reducing the likelihood of proper return and disposal of HWM [50]. It is noteworthy that the amount of medication in package contents negatively affects medical prescriptions. Controlled drugs, regulated by prescriptions (e.g., antibiotics and antidepressants), often have larger quantities in their packages than necessary for the prescribed treatment [9]. Physicians are compelled to prescribe larger amounts to ensure user access to the medication, despite the excess volume beyond the treatment requirements [48]. Consequently, the sixth hypothesis was formulated to investigate this phenomenon.

Hypothesis 6 (H6). 

The amount of medication in the package negatively moderates the medical prescription.

2.1.7. Public Policies and Laws

The establishment and enforcement of public policies and laws at national and international levels, with the aim of ensuring proper management of HWM, have a positive impact on waste reduction [51]. An example of such a policy implementation is observed in Brazil, where Law No. 17,211/2012 assigns responsibility to all stakeholders involved in the production and commercialization of medicines and their packaging to actively participate in HWM management. Pharmacies are required to provide collection spaces for EOU and/or EOL medications, manufacturers must supply containers for the collection of these medications, and public health inspection agencies are responsible for overseeing the entire process [52]. Hence, the seventh hypothesis was formulated to investigate the influence of public policies and laws on HWM management.

Hypothesis 7 (H7). 

Public policies and laws positively influence user cooperative behavior to properly manage HWM, thus avoiding pharmacopollution.

Thus, the theoretical model resulting from the development of the hypotheses is presented in Figure 1.

2.2. Research Universe and Sample

Despite the existence of public policies and laws, such as Law No. 17,211/2012, which regulates the management of HWM at the national level in Brazil, there is a lack of proactive initiatives by stakeholders such as pharmacists, pharmacies, hospitals, and medicine users to implement the provisions of the law. Consequently, the management of HWM in Brazil loses its significance and contributes to an increased risk of pharmacopollution [9]. This situation is particularly concerning in the northeastern region of Brazil, where there is a lack of incentives for user involvement in proper HWM management, thereby posing a greater threat to public health [52]. To focus our research and sample population, we specifically targeted medicine users from the state of Pernambuco in the northeast region of Brazil. Ethical approval for the study was obtained from a local ethics committee (registration number 52719621,7,0000,5206), and respondents were selected using a non-probabilistic convenience sampling approach.

2.3. Data Collection Instrument

For data collection in this study, a questionnaire was developed following the methodology proposed by [53]. The primary objective of the questionnaire was to identify creative intentional and unintentional factors influencing users’ involvement in cooperating toward managing their home waste medicine and preventing pharmacopollution.

The questionnaire was constructed based on validated constructs from the literature, which were translated into Portuguese, the first language of the researchers involved in this study. It was organized into seven main topics, and each topic represented a specific construct. The intentional factors included user attitude [36], marketing campaigns [39], package contents [50], and safe handling [46]. On the other hand, the unintentional factors encompassed medical prescription [48], collection points [44], and public policies and laws [23].

Most of the questions were closed-ended, supplemented by a small number of open-ended questions to provide additional insights. To ensure the statistical integrity of the data collection instrument, a reliability and internal consistency analysis was conducted [53]. Subsequently, the questionnaire was made available online, allowing respondents to self-administer it in a convenient manner.

2.4. Data Analysis

We used SPSS-AMOS version 27.0.0/free trial for data analysis, starting with the formulation of a Structural Equation Model (SEM) with the variables in Figure 1. The SEM technique is widely used in the literature for its advantages when compared to other techniques because (a) it analyzes the relationship between variables such as robustness, efficiency, and more significant results, (b) it can analyze more indicators of variables per construct, simultaneously, (c) it considers the measurement of variable errors, (d) the results of associations between variables are unbiased and reliable, and (e) it can test complex relationships and their associated hypotheses [31].

Also, this study used an artificial neural network (ANN), which is an artificial intelligence technique that simulates a human brain to the learning process in a dataset. We also integrated an SEM and ANN to make the intended outputs more robust. However, traditional approaches that integrate an SEM with ANN are argued to be shallow in their analysis. In this sense, the present study prompted the adoption of a deep learning SEM-ANN structure recommended by [32] aiming to explore more complex relationships among variables. A multi-layer perceptron with three interconnected layers (input, hidden, and output) was utilized, connected by synaptic weights. Non-linear activation functions, specifically the sigmoid function, were applied, and hidden neurons were generated automatically. The dataset was divided, with 90% used for training and the remaining for testing, allowing for error computation.

To mitigate overfitting, a 15-fold cross-validation procedure was implemented. The evaluation involved calculating the root mean square error (RMSE) to assess prediction quality, and a sensitivity analysis was performed to evaluate the average importance of inputs, analogous to path coefficients in regression models.

It is worth mentioning that we excluded the cases of observations with missing data as we believed the missingness to be completely random, and the resulting sample remained representative and unbiased. To this end, it was crucial to carefully assess the potential impact of excluding cases on the study’s findings by a sensitivity analysis and ensure that the excluded cases did not introduce selection bias, as suggested by [32].

3. Results and Discussion

3.1. Demographic Profile of the Respondents

A sample size of n = 952 was obtained for the study upon completion of data collection. This sample size is considered adequate for a structural equation modeling (SEM) analysis, as recommended by [33], who suggests a minimum of 200 respondents. The demographic characteristics of the respondents are presented in

Table 1. Demographic profile of the respondents. Source: the authors (2023).

Demographic Feature	Option	Responses	Percentage (%)
Gender	Female	536	56.30%
	Male	416	43.70%
Age	18–21	127	13.30%
	22–28	408	42.90%
	29–35	176	18.50%
	36–41	143	15.00%
	Above 41	98	10.30%
Income (103 reais—Brazilian currency)	Less than 1.32	147	15.40%
	1.32–3.96	290	30.50%
	3.96–6.6	195	20.50%
	6.6–13.2	173	18.20%
	13.2–26.4	109	11.40%
	Above 26.4	38	4.00%
Education	Illiterate	0	0.00%
	Primary	52	5.40%
	High school	179	18.80%
	College degree	489	51.40%
	Graduate	232	24.40%

.

From Table 1, it can be seen that the majority of respondents are female (56.30%), aged between 22 and 28 years (42.90%), self-declared as white (46.20%), have an average income of between BRL 1.32 and BRL 3.96 thousand (Brazilian currency), and hold a college degree (51.4%).

3.2. Model Fitting

To assess the adequacy of the sample, we initially conducted Bartlett’s Test of Sphericity (BTS) and the Kaiser–Meyer–Olkin (KMO) test. Both tests yielded satisfactory results, with BTS = 10,188,920 (p < 0.001) and KMO = 0.903, indicating the suitability of the sample for factor analysis. Subsequently, an Exploratory Factor Analysis (EFA) was performed to identify the underlying pattern structure, as presented in

Table 2. Exploratory Factor Analysis and Convergent Validity Test. Source: the authors (2023).

Variable	Item	Standard Loading	CR	AVE	Cronbach’s Alpha
Attitude			0.913	0.612	0.933
ATD1	I have a favorable attitude towards HWM	0.744
ATD2	I have a favorable attitude to the view that disposing of HWM adequately is a benefit	0.757
ATD3	I have a favorable attitude to the view that disposing of HWM adequately has good effects on the environment	0.720
ATD4	I have a favorable attitude to the view that disposing of HWM adequately has good effects on society	0.923
Marketing Campaigns			0.958	0.572	0.872
MKT1	I have access to marketing campaigns on the adequate management of HWM	0.907
MKT2	I know about the adequate management of HWM because of marketing campaigns	0.913
MKT3	I will certainly manage HWM adequately if the marketing campaigns show me how to do this	0.814
Collection Points			0.973	0.641	0.922
CLT1	I am familiar with where there are collection points that you can get to easily	0.721
CLT2	I know where there are collection points I can get to easily from my home	0.989
CLT3	I can easily get to collection points on my way to my place of work	0.732
Safety in Handling			0.781	0.501	0.849
SFT1	I am able to handle HWM safely	0.753
SFT2	I have total control over the handling of HWM	0.757
SFT3	It would be great if they could collect HWM from my home	0.720
Medical Prescription			0.812	0.706	0.860
PRS1	The prescription for medicines is never bigger than necessary	0.989
PRS2	I am willing to pay to purchase the amount of medication prescribed by the doctor	0.819
PRS3	I am confident in the amount of medication prescribed by the doctor	0.972
PRS4	I will take the amount of medication prescribed by the doctor if my treatment is effective	0.844
Content of the Packaging			0.953	0.533	0.963
QTY1	I sometimes buy a larger number of medications on the contents of the packaging than is on the doctor’s prescription	0.723
QTY2	I will pay to buy a larger amount of medication because of the contents declared on the packaging	0.714
QTY3	I buy a larger amount of medication when the sales team offers me benefits because of the contents present on the packaging	0.734
Public Policies and Laws			0.873	0.674	0.892
PPL1	I am familiar with public policy and law regarding the proper management of HWM	0.713
PPL2	I intend to dispose of HWM properly if public policy and laws will be more restrictive	0.726
Cooperative Behavior
CB1	My behavior as to properly managing HWM is cooperative when I know what the risks involved are	0.718	0.832	0.553	0.829
CB2	I behave cooperatively as to properly managing HWM when I know what the public policies and laws that regulate this activity are	0.837
CB3	I will manage HWM properly because it will be good for society and the environment	0.824

[34].

To ensure the internal consistency of the variables, we employed the Composite Reliability (CR) analysis. The results showed that all variables exhibited CR values greater than 0.7, indicating good consistency. Furthermore, the Convergent Validity test was conducted to assess the theoretical interconnectedness of the items using Average Variance Extracted (AVE) and Standard Loadings. All variables achieved AVE values exceeding 0.5, indicating that the latent variables of the model captured at least 50.0% of the variance. Additionally, the Standard Loadings were higher than 0.7, further confirming the validity of the model. Refer to Table 2 for details [33].

For reliability analysis, we employed Cronbach’s Alpha and CR measures. All variables yielded values between 0.8 and 0.9, indicating high reliability. Moreover, the CR values were greater than 0.7 (see Table 2), further supporting the reliability of the variables, as these values are widely accepted in the literature [54].

Discriminant validity was evaluated using the Heterotrait–Monotrait (HTMT) criterion, and the results are presented in

Table 3. Factor correlations and discriminant validity (HTMT_0.85 criterion). Source: the authors (2023).

Variable	ATD	MKT	CLT	SFT	PRS	QTY	PPL
ATD	[0.782] *
MKT	0.432	[0.756] *
CLT	0.283	0.472	[0.801] *
SFT	0.547	0.274	0.423	[0.708] *
PRS	0.387	0.343	0.264	0.527	[0.840] *
QTY	0.102	0.641	0.511	0.321	0.320	[0.730] *
PPL	0.397	0.467	0.605	0.432	0.220	0.764	[0.821] *

* Values in italics are the square root of the AVE.

. All values obtained were below the threshold of 0.85, indicating satisfactory discriminant validity. Additionally, the square root of all Average Variance Extracted (AVE) values exceeded their respective correlations with other variables, further confirming the discriminant validity of the measurement model [33].

3.3. Estimations of Structural Equation and Hypotheses Testing

A Confirmatory Factor Analysis (CFA) was conducted to assess the fit of the data to the proposed model. Various goodness-of-fit tests were performed, including the Chi-square degree of freedom ratio (CMIN/df), Comparative Fitting Index (CFI), Tucker–Lewis Index (TLI), Goodness-of-fit Index (GFI), Relative Fit Index (RFI), Normed Fit Index (NFI), Incremental Fit Index (IFI), and root mean square of the similarity error (RMSEA). The obtained values (CMIN/df = 1.618, CFI = 0.926, TLI = 0.972, GFI = 0.912, RFI = 0.888, NFI = 0.897, IFI = 0.894, and RMSEA = 0.069) were in accordance with the literature, indicating a good fit of the model [55].

For hypothesis testing and estimation in Structural Equation Modeling (SEM), the coefficient of determination (R²) was calculated, which measures the proportion of variation in the dependent variable explained by the independent variables. The obtained R² value of 0.64 surpassed the threshold of R² = 0.35 suggested by [56], indicating a significant interpretation of the model. Furthermore, multicollinearity was evaluated, and no issues were found, as the Variance Inflation Factor (VIF) for all variables was below 10 [57].

Path analysis was performed for the entire SEM, and various goodness-of-fit tests were conducted (CMIN/df = 1.816, CFI = 0.936, TLI = 0.941, AGFI = 0.970, RFI = 0.913, NFI = 0.958, PCLOSE = 0.894, and RMSEA = 0.076). The results met the reference values, indicating a good fit of the model to the data [55].

In

Table 4. Hypotheses testing and path analysis. Source: the authors (2023).

Hypothesis	Hypothesized Path	Path Coefficient (β)	p-Value *	Decision	VIF	R2
H1	ATD$\to$CB	1.562	0.0700	Accept	1.381	0.64
H2a	MKT$\to$CB	0.362	0.0300	Accept	1.416
H3	CLT$\to$CB	0.552	0.0400	Accept	1.361
H4	SFT$\to$CB	−0.090	<0.0001	Reject	1.245
H5	PRS$\to$CB	0.321	0.0018	Reject	1.323
H7	PPL$\to$CB	0.164	0.0150	Accept	1.056
Moderating Effect
H2b	MKT$\to$ATD	0.591	0.0700	Accept	1.321
H6	QTY$\to$PRS	−0.336	0.0300	Accept	1.116

* α = 0.01.

, the hypotheses were tested using p-values, and it was observed that only Hypotheses H4 and H5 lacked evidence to support them.

Hypothesis H4, which examines the relationship between safe handling and cooperative behavior, was rejected (p-value = <0.0001, β = −0.090), aligning with findings in the literature [46]. However, this result should be analyzed considering users’ concerns about environmental contamination rather than personal safety when handling waste. Users may fear that improper handling during waste disposal could lead to environmental pollution due to their awareness of associated risks. Implementing door-to-door collection programs could help alleviate this fear and promote medicine users’ involvement in cooperating with HWM management [47].

Similarly, Hypothesis H5, which explores the relationship between medical prescription and cooperative behavior, was rejected (p-value = 0.0018, β = 0.321), which is also in accordance with the existing literature [44]. It is crucial to evaluate this result within the context of Brazil as a developing country. Controlled medications are only distributed through the Unified Health System in exact quantities prescribed by physicians. Therefore, cooperative behavior is not influenced by prescription requirements since the necessity of continuing treatment is assessed, and medications may be discontinued if deemed unnecessary [48,49].

Hypothesis H1, examining the relationship between attitude and cooperative behavior, was supported (p-value = 0.0700, β = 1.562). Researchers suggest that users’ cooperative behavior in HWM to prevent pharmacopollution is influenced by their evaluation of optimistic and pessimistic opinions regarding ecological attitudes [58]. Optimistic opinions may include the long-term health benefits for their community and the preservation of scarce consumable water sources, particularly relevant in the northeastern region of the study. Pessimistic opinions may focus on the economic aspect, where unused medications are seen as wasted investments [52,59]. Such analyses can shape users’ attitudes toward cooperative behavior and also their involvement in HWM management.

Hypothesis H2a, exploring the relationship between marketing and cooperative behavior, was statistically supported (p-value = 0.0300, β = 0.362). The literature emphasizes the importance of raising users’ awareness through marketing campaigns, including social media and other communication channels, to promote proper medication usage. These campaigns should also highlight the benefits of reducing HWM generation, as increased awareness positively impacts users’ cooperative behavior by discouraging inappropriate disposal practices such as discarding medications in the trash or down the drain.

Hypothesis H3, investigating the relationship between collection points and cooperative behavior, was also accepted (p-value = 0.0400, β = 0.552). According to [60], users are more likely to be involved and cooperate in proper HWM at collection points when they perceive convenience in terms of accessibility, including factors such as time, location, and proximity. Even users with a strong awareness of appropriate HWM may not involve themselves in HWM management with cooperative behavior if access to collection points is challenging. Therefore, the spatial distribution of collection points should be designed to maximize proximity to users and facilitate their utilization.

Furthermore, Hypothesis H7, exploring the relationship between public policy and law and cooperative behavior, was supported (p-value = 0.0150, β = 0.164). In Brazil, the relationship between public policies, laws, and user cooperation presents challenges, as existing initiatives at the national level do not effectively encourage users to increase their medication return rates [23]. However, adopting a strategy of mutual benefits that aligns with user needs and compliance with regulations can enhance return rates and promote cooperative behavior [61].

The study also tested the proposed moderating effects in Hypotheses H2b and H6. The moderating effect of H2b, which investigates the relationship between marketing and user attitude, was statistically supported (p-value = 0.0700, β = 0.591). This highlights the importance of promoting green marketing practices and their associated benefits to address inaction and shape consumers’ attitudes. Authorities, businesses, and conscious consumers should collaborate to foster such practices [62].

Similarly, the moderating effect in H6, which examines the relationship between package contents and medical prescription, was supported by statistical evidence from the sample (p-value = 0.0300, β = −0.336). Persuasion by sales teams to encourage users to purchase larger quantities of medication than prescribed contributes to increased sales but also leads to higher levels of HWM [47]. Users tend to store the excess medication, often without using it, exacerbating the mismanagement of HWM and contributing to pharmacopollution [63].

3.4. Estimations of Artificial Neural Networks Models

Moreover, regarding the ANN analysis, according to the theoretical model proposed in Figure 1, we developed three ANN models, as depicted in Figure 2. In Model 1 (Figure 2a), the dependent variable is CB, while in Models 2 and 3 (Figure 2b,c), the dependent variables are ATD and PRS, respectively. It is important to emphasize that Model 1 represents the direct relationships between variables in the SEM, whereas Models 2 and 3 represent the moderating effects of the SEM.

After conducting cross-validation, the results are summarized in

Table 5. Root mean square error (RMSE). Source: the authors (2023).

Model 1: Training	N	SSE	RMSE	Model 1: Testing	N	SSE	RMSE
ANN1	857	12.159	0.126	ANN1	95	0.857	0.119
ANN2	849	11.588	0.128	ANN2	103	1.416	0.138
ANN3	847	11.520	0.129	ANN3	105	1.261	0.124
ANN4	854	10.728	0.130	ANN4	98	1.345	0.122
ANN5	856	11.117	0.144	ANN5	96	1.223	0.125
ANN6	848	11.502	0.145	ANN6	104	1.281	0.138
ANN7	850	11.422	0.127	ANN7	102	1.445	0.127
ANN8	852	10.222	0.138	ANN8	100	1.523	0.132
ANN9	854	10.785	0.144	ANN9	98	1.656	0.136
ANN10	855	11.142	0.143	ANN10	97	1.785	0.132
ANN11	856	10.532	0.129	ANN11	96	1.822	0.134
ANN12	857	10.012	0.132	ANN12	95	1.433	0.135
ANN13	855	11.152	0.154	ANN13	97	1.525	0.131
ANN14	853	10.789	0.136	ANN14	99	1.322	0.128
ANN15	852	11.123	0.133	ANN15	100	1.454	0.127
		Mean	0.126			Mean	0.129
		SD	0.008			SD	0.006
Model 2: Training	N	SSE	RMSE	Model 2: Testing	N	SSE	RMSE
ANN1	856	16.159	0.146	ANN1	96	1.481	0.107
ANN2	850	14.588	0.140	ANN2	102	1.516	0.128
ANN3	848	13.520	0.159	ANN3	104	1.461	0.104
ANN4	855	11.728	0.140	ANN4	97	1.345	0.123
ANN5	853	12.117	0.143	ANN5	99	1.423	0.124
ANN6	849	13.502	0.146	ANN6	103	1.281	0.135
ANN7	852	14.422	0.147	ANN7	100	1.445	0.122
ANN8	853	15.222	0.148	ANN8	99	1.423	0.122
ANN9	854	13.785	0.145	ANN9	98	1.156	0.137
ANN10	855	12.142	0.142	ANN10	97	1.252	0.135
ANN11	856	11.532	0.152	ANN11	96	1.353	0.136
ANN12	857	12.012	0.142	ANN12	95	1.444	0.133
ANN13	855	12.152	0.153	ANN13	97	1.445	0.135
ANN14	853	11.789	0.152	ANN14	99	1.246	0.128
ANN15	852	12.123	0.136	ANN15	100	1.237	0.127
		Mean	0.146			Mean	0.126
		SD	0.006			SD	0.010
Model 3: Training	N	SSE	RMSE	Model 3: Testing	N	SSE	RMSE
ANN1	855	15.559	0.136	ANN1	97	1.481	0.139
ANN2	852	13.488	0.150	ANN2	100	1.516	0.136
ANN3	848	12.320	0.149	ANN3	105	1.461	0.134
ANN4	853	13.428	0.150	ANN4	104	1.345	0.132
ANN5	852	14.317	0.153	ANN5	100	1.423	0.135
ANN6	849	12.302	0.136	ANN6	103	1.281	0.138
ANN7	851	15.322	0.147	ANN7	101	1.445	0.127
ANN8	854	16.422	0.158	ANN8	98	1.423	0.134
ANN9	854	14.685	0.135	ANN9	98	1.156	0.135
ANN10	855	13.442	0.152	ANN10	97	1.252	0.136
ANN11	856	12.332	0.142	ANN11	96	1.353	0.134
ANN12	857	13.112	0.152	ANN12	95	1.444	0.136
ANN13	855	14.252	0.143	ANN13	97	1.445	0.132
ANN14	853	12.489	0.155	ANN14	99	1.246	0.138
ANN15	852	15.223	0.156	ANN15	100	1.237	0.139
		Mean	0.148			Mean	0.135
		SD	0.008			SD	0.003

.

From Table 5, the average RMSE values for Model 1 (training = 0.126, testing = 0.129), Model 2 (training = 0.146, testing = 0.126), and Model 3 (training = 0.148, testing = 0.135) align with established quality metrics in the literature.

Furthermore, a sensitivity analysis was performed to determine the average and normalized importance of inputs for Models 1, 2, and 3, as shown in

Table 6. Sensitivity analysis. Source: The authors (2023).

Model 1: Output = CB	ATD	MKT	CLT	SFT	PRS	PPL	Model 2: Output = MKTATD	Model 3: Output = QTY PRS
ANN1	0.232	0.143	0.146	0.012	0.014	0.113	1	1
ANN2	0.259	0.102	0.148	0.013	0.015	0.112	1	1
ANN3	0.257	0.132	0.149	0.013	0.023	0.122	1	1
ANN4	0.231	0.122	0.150	0.014	0.024	0.103	1	1
ANN5	0.243	0.125	0.154	0.016	0.106	0.105	1	1
ANN6	0.282	0.126	0.155	0.017	0.116	0.116	1	1
ANN7	0.285	0.124	0.157	0.010	0.015	0.114	1	1
ANN8	0.284	0.125	0.148	0.007	0.008	0.106	1	1
ANN9	0.276	0.121	0.154	0.021	0.024	0.123	1	1
ANN10	0.263	0.127	0.153	0.026	0.036	0.120	1	1
ANN11	0.238	0.128	0.149	0.030	0.105	0.124	1	1
ANN12	0.240	0.129	0.142	0.012	0.106	0.112	1	1
ANN13	0.248	0.124	0.154	0.017	0.108	0.114	1	1
ANN14	0.243	0.123	0.146	0.013	0.110	0.111	1	1
ANN15	0.244	0.120	0.153	0.060	0.109	0.110	1	1
Average Importance	0.232	0.143	0.146	0.012	0.014	0.113	1	1
Normalized Importance (%)	100.000	61.637	62.931	5.172	6.034	48.706	100.000	100.000

.

In Model 1, the highest predictive value was observed for ATD (100.000%), followed by CLT (62.932%), MKT (61.637%), PPL (48.706%), PRS (6.034%), and SFT (5.172%). Notably, the ANN model yielded a similar input importance ranking to that of the significant independent variables in the SEM model, as displayed in

Table 7. Comparing results of SEM versus ANN (dependent variable = CB). Source: the authors (2023).

	SEM			ANN
Independent Variable	Path Coefficient	Result	Ranking as per Path Coefficient	Ranking	Normalized Importance (%)	Remark
ATD	1.562	Significant	1	1	100.000	Matched
MKT	0.362	Significant	3	3	61.637	Matched
CLT	0.552	Significant	2	2	62.931	Matched
SFT	−0.090	Non-Significant	-	6	5.172
PRS	0.321	Non-Significant	-	5	6.034
PPL	0.164	Significant	4	4	48.706	Matched

.

It is worth mentioning that, despite SFT and PRS not being significant in the SEM model, they had relative importance for predicting the CB in the ANN model, even with lower values compared to other inputs. This discrepancy arises from the ANN algorithm’s ability to capture non-linear relationships between variables, offering significant advancements in analysis. The ANN also demonstrated its outstanding contribution by effectively predicting the moderating effects of SEM in Models 2 and 3, particularly for ATD and PRS. By performing a dual-stage SEM-ANN analysis based on deep learning algorithms, we ensure a comprehensive exploration of the relationships between various factors. Omitting the two hidden layers in the deep artificial neural network architecture may lead to a potential oversight of the true impact of SFT and PRS on CB. These results are aligned with the results of [32], which proposed the algorithms used in this study.

4. Conclusions

This study aimed to identify creative intentional and unintentional factors influencing the involvement of medication users to cooperatively behave in managing their waste to prevent pharmacopollution. Seven factors were identified (i.e., attitude, marketing campaigns, collection points, safe handling, medical prescription, package contents, and public policies and laws), and a deep learning SEM-ANN was developed to test the relationships between these variables. The analysis confirmed that the SEM-ANN effectively achieved its objective.

The findings revealed that among the intentional factors such as attitude, marketing campaigns, package contents, and safe handling, only the hypothesis regarding safe handling and cooperative behavior (H4) was not supported. This indicates that respondents were more concerned about the environmental risks associated with improper waste handling rather than their personal safety when disposing of medication waste, considering that the medications had already been used for treatment.

Regarding the unintentional factors, including medical prescription, collection points, and public policies and laws, only the hypothesis linking medical prescription to cooperative behavior (H5) was not supported. This result primarily pertains to controlled prescription drugs, where medications can only be obtained in the prescribed quantities for treatment purposes. When comparing SEM and ANN results, we could observe similar rankings regarding the inputs used to predict the outputs. Moreover, the ANN was able to capture non-linear relationships among variables that were not observed in the SEM result, thus, contributing to our analysis.

In addition to contributing to the existing literature on the topic, this study offers practical implications. It calls for further investigation and the development of practices to promote cooperative behavior among the involvement of medication users, emphasizing the proper use and disposal of medications to prevent pharmacopollution. Long-term studies are also encouraged to explore the replacement of environmentally toxic medications with alternatives that have a minimal environmental impact. Furthermore, the study urges researchers to develop medications that degrade more rapidly and minimize contamination levels, facilitating their removal from the environment with fewer adverse effects.

Policy-makers are urged to take a more proactive role in interventions by formulating public policies and enacting laws that consider the perspective of medication users. Increased utilization of supervisory and regulatory bodies is recommended to ensure industry and commerce fulfill their responsibilities in managing medication waste beyond their business premises. Engaging in dialogues with users, industry stakeholders, and other relevant parties to establish a shared vision for HWM management is crucial.

For the pharmaceutical industry and medication retailers, it is important to reflect on the balance between financial profitability and the potential negative consequences of indiscriminate medication sales. Guidance should be provided to sales staff to promote increased sales without compromising public well-being.

Society as a whole needs to reconsider its consumption habits, including implementing simple mechanisms to monitor personal medication stocks and actively participating in proper disposal practices, such as utilizing collection points or supporting door-to-door collection programs.

While this study has limitations, these limitations also serve as suggestions for future research. Replicating the study in rural areas of the state of Pernambuco would help determine if the results can be generalized beyond the metropolitan region. Additionally, developing decision support models to assist public and private managers in prioritizing actions to mitigate the impact of pharmacopollution resulting from improper HWM would be valuable. We plan to use advanced optimization algorithms, including hybrid heuristics and metaheuristics, adaptive algorithms, self-adaptive algorithms, and island algorithms, that have gained significant attention in solving challenging decision problems [64,65,66,67,68,69]. In the context of prioritizing actions to mitigate the impact of pharmacopollution resulting from improper household waste medicine (HWM) management, exploring these advanced optimization algorithms holds promise for enhancing the effectiveness of decision-making processes.

The importance of advanced optimization algorithms lies in their ability to tackle complex decision problems by efficiently searching for optimal or near-optimal solutions in large solution spaces. These algorithms incorporate sophisticated techniques inspired by natural processes, computational intelligence, and mathematical optimization, allowing them to address problems with multiple constraints, conflicting objectives, and uncertainties [64,65,66,67,68,69].

Notably, advanced optimization algorithms have been successfully applied across various domains, extending beyond the specific decision problem addressed in this study. Domains such as online learning, scheduling, multi-objective optimization, transportation, medicine, and data classification have benefited from these algorithms’ solution approaches. They have been instrumental in optimizing resource allocation, minimizing costs, improving system efficiency, and aiding decision-making processes [64,65,66,67,68,69].

For the decision problem of prioritizing actions to mitigate the impact of pharmacopollution resulting from improper HWM, advanced optimization algorithms can offer several potential applications. These algorithms can aid in identifying optimal action plans, allocating limited resources effectively, and minimizing the negative consequences of pharmacopollution. By incorporating factors such as geographical distribution, resource availability, waste disposal methods, and environmental impact, these algorithms can provide valuable insights for decision makers.

In future research, it would be beneficial to compare the proposed approach in this study with advanced heuristic and metaheuristic algorithms to assess their relative performance and efficiency. By conducting such comparisons, researchers can evaluate the strengths and limitations of different optimization techniques and identify the most effective strategies for prioritizing actions and mitigating pharmacopollution resulting from improper HWM.