Modified Laboratory Risk Indicator and Machine Learning in Classifying Necrotizing Fasciitis from Cellulitis Patients

Abstract

Necrotizing fasciitis (NF) is a severe and life-threatening soft tissue infection that requires timely and accurate diagnosis to improve patient outcomes. The early diagnosis of NF remains challenging due to its similarity to other subcutaneous soft tissue infections like cellulitis. This study aims to employ machine learning techniques to differentiate NF from cellulitis and enhance the diagnostic accuracy of NF by developing a modified LRINEC (MLRINEC) score. These modifications aimed to improve the sensitivity and specificity of NF diagnosis. The study utilized three machine learning classifiers—Logistic Regression, decision tree, and Random Forest—to assess their effectiveness in distinguishing between NF and cellulitis cases. The MLRINEC score was developed by incorporating six key blood test parameters: creatinine, hemoglobin, platelet count, sodium, white blood cell count, and C-reactive protein using laboratory data from Maha Sarakham Hospital in Northeastern Thailand. Our findings indicate that the decision tree classifier demonstrated superior performance, achieving the highest recall, particularly in accurately identifying NF cases. A feature importance analysis revealed that hemoglobin levels and white blood cell counts were the most critical factors influencing the model’s predictions. The platelet count (PT), C-reactive protein (CRP), and creatinine (CT) also played important roles, while sodium levels (NA) were the least influential. The MLRINEC score demonstrates high accuracy in classifying NF and cellulitis patients, paving the way for improved diagnostic protocols in clinical settings.

1. Introduction

Necrotizing fasciitis (NF) is a profound inflammation of soft tissue within the fascia stemming from a severe infection, predominantly bacterial in nature. Surgical intervention may be necessary for localized infections, particularly in extremities. Delays in treatment and inappropriate antibiotic use significantly heighten mortality risks for affected individuals. Diagnosis to separate necrotizing fasciitis from other subcutaneous soft tissue infections that do not require surgery such as cellulitis is difficult in the early stages. The readily available laboratory biomarkers from routine blood and biochemistry examinations leading to the risk prediction model can be an ideal tool for screening or ruling out NF owing to their low cost, rapid access, and availability.

Machine learning (ML) is increasingly valuable in medical recommendations, aiding in disease diagnosis and risk factor identification. The application of ML models for early diagnosis and prognosis significantly enhances disease management. By enabling precise and timely disease prediction, ML allows physicians to implement the most effective treatments, thereby minimizing diagnostic uncertainty. For example, Chia-Peng. et al. (2024) [1] identified necrotizing soft tissue infection using infectious fluid analysis and clinical parameters based on machine learning algorithms. The research was concluded that machine learning algorithms significantly improve the identification of necrotizing soft tissue infections. By utilizing infectious fluid analysis alongside clinical parameters, these algorithms can provide a more accurate diagnosis, which is critical for timely intervention and treatment. Additionally, the study highlighted that the use of AI and machine learning minimizes biases inherent in traditional diagnostic methods, leading to greater precision in diagnosing necrotizing soft tissue infections—an essential factor for effective patient management. Moreover, Kinnor, Das. (2024) [2] emphasizes the transformative role of AI in medical diagnostics, detailing its applications, benefits, and the challenges that need to be addressed for successful implementation. In this study, we aim to evaluate its effectiveness in differentiating necrotizing fasciitis (NF) from other soft tissue infections using machine learning techniques, with data sourced from Maha Sarakham Hospital in Northeastern Thailand. This region has a notably high prevalence of patients affected by the condition, with incidence peaking during the rainy season. The disease predominantly impacts individuals with predisposing factors, such as existing wounds or congenital conditions, and those exposed to the pathogen. Farmers, a significant portion of the local population, are particularly affected.

In 2004, Wong et al. [3] introduced the Laboratory Risk Indicator for Necrotizing Fasciitis (LRINEC) score, designed to evaluate patients exhibiting symptoms suggestive of subcutaneous tissue infection with the aim of gauging the risk of necrotizing fasciitis (NF). This scoring system incorporates six common serum parameters: C-reactive protein, total white cell count, hemoglobin, sodium, creatinine, and glucose. Subsequent research, such as that by Thomas et al. in 2015 [4], refined the LRINEC score by modifying laboratory parameters and incorporating clinical indicators, resulting in significant enhancements to the scoring system. Their findings highlighted the significance of C-reactive protein values while suggesting that serum sodium and glucose levels offer limited value. They proposed replacing glucose and sodium levels with erythrocyte count and fibrinogen levels to improve the accuracy of the score. Subsequently, El-Menyar (2017) [5] investigated a retrospective analysis for patients who were admitted with NF at Hamad General Hospital (HGH) based on standard LRINEC points. Patients were classified into (Group 1: LRINEC < 6 and Group 2: LRINEC ≥ 6). The two groups were analyzed and compared. Next, Wu et al. 2021 [6] added serum lactate and comorbid liver disease to the original LRINEC score and re-defined the cut-off values for three variables to develop the MLRINEC score. Logistics regression was used to determine the association with NF after adjustment for confounders and the MLRINEC score was developed by them. Moreover, there were many researchers investigating for finding an effective diagnostic tool to develop the LRINEC score. However, the LRINEC score is a sensitive diagnostic tool, but it is less specific.

In Pathophysiology, disseminated intravascular coagulation (DIC) can occur in severe NF, where there is a widespread activation of the coagulation cascade. This leads to the formation of clots throughout the small blood vessels, depleting fibrinogen and platelets and causing bleeding and organ dysfunction. Utilizing fibrinogen and platelets to diagnose NF by considering DIC is associated with NF by fibrinogen levels which may drop due to consumption, and platelet counts are also reduced. Thrombocytopenia (low platelet count) can be observed in NF patients and may be indicative of severe systemic inflammation or disseminated intravascular coagulation (DIC), which can occur in severe infections. Increased levels of fibrin degradation products like D-dimer are also indicative of DIC. Both fibrinogen and platelets contribute to the formation of microthrombi in the blood vessels within the affected tissue. This can lead to further tissue ischemia and damage, exacerbating the disease. Incorporating platelet counts into diagnostic models, like the modified LRINEC score, could enhance the accuracy of distinguishing NF from other soft tissue infections.

Wong et al. [3] introduced a classical LRINEC score that incorporates six parameters from a blood test, while Thomas et al. (2015) [4] suggested improving accuracy by replacing glucose and sodium levels with erythrocyte count and fibrinogen levels. Therefore, in this study, we further adapt the LRINEC score by substituting platelet counts for glucose levels used in the classical LRINEC score. We modified it by incorporating the six laboratory blood tests: creatinine (CT), hemoglobin (HB), platelet count (PT), sodium (NA), white blood cells (WBCs), and C-reactive protein (CRP). The findings of this study can be applied to diagnose and treat necrotizing fasciitis (NF) more effectively, thereby reducing the severity of the disease in affected patients.

2. Materials and Methods

2.1. Dataset

The dataset for this study includes inpatient records from Maha Sarakham Hospital spanning 2014 to 2020. Patients were identified using ICD-10 codes M7260-M7269 for necrotizing fasciitis (NF) and L030-L039 for cellulitis. Out of these records, 144 patients had complete data for all six lymphatic diagnostic parameters. The CRP values were assessed using two methods: CRP-Nephelometry, which provides a quantitative measurement, and CRP-C-Reactive Protein, which yields a qualitative positive or negative result. To ensure consistency, the CRP values above 150 mg/L were recoded as “Positive”, while those below were recoded as “Negative” .

2.2. Classification Methods

Our study seeks to differentiate between the patients with suspected necrotizing fasciitis (NF) and those with cellulitis using supervised classification techniques. We train the model with input variables to enable it to accurately classify cases based on learned data. We selected three classifiers for this analysis, chosen for their suitability given the dataset size, the presence of a labeled training dataset, and their prevalence in similar research. Each classifier is detailed in the following subsections.

2.2.1. Logistic Regression

Logistic Regression is commonly used for binary classification tasks such as spam detection and disease diagnosis (predicting if a patient has NF or not). The model’s output is a probability value between 0 and 1, which is obtained using the logistic function (also known as the sigmoid function). The logistic function is defined as follows:

$$P(\mathrm{Y}=1|X)={\displaystyle \frac{e^{{\beta }_0+{\beta }_1+x_1+\dots +{\beta }_k{x}_k}}{1+e^{{\beta }_0+{\beta }_1{x}_1+\dots +{\beta }_k{x}_k}}};$$

where ${\beta }_0$ is the intercept, ${\beta }_i$ are the coefficients, and $x_i$ are the predictor variables; $P(\mathrm{Y}=1|X)$| defined the probability of having NF when determining the values of the predictive variables.

2.2.2. Decision Tree Classifier

The decision tree (DT) classifier [7] is a versatile algorithm that structures a dataset into a tree-like model based on various features. This hierarchical approach allows the data to be split at multiple levels, enabling the efficient handling of large volumes of information. The decision tree method involves constructing nodes and branches that represent different features and their corresponding test cases. Each node in the tree evaluates a specific feature, with branches representing the possible outcomes of the test. This recursive process continues for each subsequent subtree. Decision trees utilize metrics such as information gain to determine the optimal way to split nodes, enhancing the model’s effectiveness for both regression and classification tasks.

2.2.3. Random Forest

Random Forest (RF) [8] is a powerful ensemble learning technique that combines multiple randomized decision trees to make predictions based on a given set of input data. This method is renowned for its accuracy in building predictive models and assessing the significance of individual features. Additionally, RF is highly effective in managing large datasets and handling biased data, demonstrating its versatility across various machine learning applications.

2.3. Program for Analysis

For classification analysis, we utilized Python version 3.10 and Scikit-learn [9]. The dataset was used for training with 10-fold cross-validation, while the test set, comprising 10% of unseen data, was reserved for evaluation.

2.4. Metrics of Model Evaluation

Classification models are used to predict the target class of a dataset by estimating the probability that each sample belongs to a specific class. Evaluating the performance of these models is essential for addressing real-world problems. In machine learning, various metrics are employed to assess the effectiveness of classification models.

Table 1. Confusion metric.

		Predicted Result		Sum of Actual Result
		Positive (Necrotizing Fasciitis: NF)	Negative (Cellulitis)
Actual Condition	Positive (Necrotizing Fasciitis: NF)	TP	FN	TP + FN
	Negative (Cellulitis)	FP	TN	FP + TN
Sum of prediction result		TP + FP	FN + TN	TP + FN + FP + TN

presents the confusion matrix, illustrating the performance of the classification model by detailing the counts of true positives (TP), false negatives (FN), false positives (FP), and true negatives (TN).

The performance of classification models is evaluated using various metrics including accuracy, precision, recall, F-measure, and specificity. These metrics are derived from the cells of the confusion matrix, and their respective formulas for calculation are provided below.

• Accuracy (AC) is a key metric that measures the proportion of correct predictions out of the total predictions made by the model.

$$\mathrm{AC}={\displaystyle \frac{\mathrm{TP}+\mathrm{TN}}{\mathrm{T}\mathrm{N}+\mathrm{FN}+\mathrm{FP}+\mathrm{TP}}}$$

• Precision (PR) is the ratio of true positive predictions to the total positive predictions made by the model, reflecting the model’s ability to minimize false positives.

  $$\mathrm{PR}={\displaystyle \frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FP}}}$$

• Recall (RC), also known as sensitivity or true positive rate, measures the proportion of actual positives correctly identified by the model, reflecting its ability to detect positive instances.

$$\mathrm{R}\mathrm{C}={\displaystyle \frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{N}}}$$

• F-measure (F1) is the harmonic mean of precision and recall, offering a balanced metric that reflects the model’s overall accuracy.

$$\mathrm{F}1=2\times {\displaystyle \frac{\mathrm{P}\mathrm{R}\times \mathrm{R}\mathrm{C}}{\mathrm{P}\mathrm{R}+\mathrm{R}\mathrm{C}}}$$

• Specificity (True Negative Rate) measures the proportion of actual negatives correctly identified by the model, highlighting its ability to minimize false negatives.

$$\mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}={\displaystyle \frac{\mathrm{T}\mathrm{N}}{\mathrm{T}\mathrm{N}+\mathrm{F}\mathrm{P}}}$$

3. Results

We will present the result of the research in two parts. That is, the first part will be the result of classifying necrotizing fasciitis and cellulitis patients using machine learning techniques that will find the best model to classify these two diseases. Part 2 will be the result of creating modify LRINEC score and present the results of its accuracy.

3.1. Machine Learning in Classifying Necrotizing Fasciitis from Cellulitis Patients

The dataset of this study has six independent variables: serum creatinine (CT), hemoglobin (HB), platelet count (PT), serum sodium (NA), white blood cell count (WBC), and C-reactive protein (CRP). All the variables are numeric values, while the CRP is in discrete value (positive and negative) and numeric form. If a numeric value is higher than 150 mg/L, it is defined as “Positive”; otherwise, it is “Negative”. The target values (dependent variable) of the study are divided into two categories: NF and non-NF.

Table 2. Summary statistics of feature variables.

Independent Variables	Dependent Variables	Range	Mean	SD	Normal Value Ranges
CT: creatinine (mg/dL)	All	0.4–13.5	1.71	2.58	male: 0.6–1.2 female: 0.5–1.1
	NF	0.4–12.4	1.49	1.97
	Non-NF	0.4–13.5	1.95	3.07
HB: hemoglobin (g/dL)	All	3.8–15.8	10.32	2.26	male: 14–18 female: 12–16
	NF	4.9–14.6	9.80	1.73
	Non-NF	3.8–15.8	10.88	2.60
PT: platelet count (cells/mm3)	All	14,000–842,000	357,444.44	166,993.21	150,000–440,000
	NF	14,000–842,000	394,148.65	180,543.75
	Non-NF	45,000–786,000	318,642.86	141,345.58
NA: sodium (mmol/L)	All	49–143	135.08	8.45	136–144
	NF	49–143	134.00	10.92
	Non-NF	122–143	136.23	4.27
WBC: white blood cells (cell/mm3)	All	3610–41,540	12,899.25	7578.10	4500–10,000
	NF	4370–41,540	13,162.03	6308.06
	Non-NF	3610–27,210	10,780.57	4812.59
CRP: C-reactive protein (mg/L)	All	Negative = 70, Positive = 74	-		<10 mg/L
	NF	Negative = 36, Positive = 38	-
	Non-NF	Negative = 34, Positive = 36	-

presents the statistical data on various blood test measurements, comparing individuals with a condition labeled “NF” (potentially a specific condition group) to those without it. Key findings include the following:

Creatinine: NF individuals generally have slightly lower coagulation test (CT) values than those without NF.

Hemoglobin Levels: Hemoglobin levels are lower on average in NF individuals compared to non-NF individuals.

Platelet Counts: Platelet counts are higher in NF individuals.

Sodium Levels: Sodium levels are slightly lower in NF individuals.

White Blood Cell Counts: NF individuals have higher white blood cell (WBC) counts, which may indicate an inflammatory or immune response.

C-Reactive Protein (CRP): The CRP data are presented as the counts of individuals with either negative or positive results, rather than as continuous measurements.

Table 3. Confusion matrix of the test set.

Model	Predicted	Actual
		Positive (Necrotizing Fasciitis: NF)	Negative (Cellulitis)
LR: Logistic Regression	Positive (Necrotizing Fasciitis: NF)	5	1
	Negative (Cellulitis)	4	5
DT: Decision Tree Classifier	Positive (Necrotizing Fasciitis: NF)	9	3
	Negative (Cellulitis)	0	3
RF: Random Forest	Positive (Necrotizing Fasciitis: NF)	8	4
	Negative (Cellulitis)	1	2

presents the confusion matrix data for the three classification models—Logistic Regression (LR), decision tree classifier (DT), and Random Forest (RF)—assessed on a test set with two classes: positive (necrotizing fasciitis) and negative (cellulitis).

Logistic Regression (LR) shows a balanced number of true positives and true negatives but also exhibits a relatively high rate of false positives and false negatives.

Decision tree classifier (DT) excels with the highest number of true positives and no false positives, though it does have some false negatives, indicating it is highly effective at identifying necrotizing fasciitis while minimizing false positives.

Random Forest (RF) demonstrates a good number of true positives but also has some false positives and false negatives, reflecting moderate overall performance.

Among the models, the decision tree classifier (DT) performs the best in accurately identifying positive cases (necrotizing fasciitis) without misclassifying negative cases (cellulitis) as positive. Logistic Regression (LR) and Random Forest (RF) each have trade-offs between false positives and false negatives, with Random Forest slightly better at minimizing false positives but performing worse than decision tree on false negatives.

Table 4. Evaluation results (%).

Model	Accuracy (AC)	Precision (PR)	Recall (RC) True Positive Rate	F-Measure (F1)	Specificity (True Negative Rate)
LR: Logistic Regression	66.67	83.33	55.56	66.67	83.33
DT: Decision Tree	80.00	75.00	100.00	85.71	50.00
RF: Random Forest	66.67	66.67	88.89	76.19	33.33

presents the evaluation results of the three classification models—Logistic Regression (LR), decision tree (DT), and Random Forest (RF)—expressed as percentages.

Decision tree (DT) shows the best performance in recall and F-measure, making it highly effective at identifying positive instances but at the cost of lower specificity.

Logistic Regression (LR) provides high precision and specificity, but its recall and F-measure are moderate.

Random Forest (RF) has good recall and a balanced F-measure but lower specificity, showing it is effective at identifying positives while struggling more with negatives.

These results suggest that the best model depends on the specific application and whether prioritizing recall (sensitivity to detecting positives) or specificity (accuracy in identifying negatives) is more critical. In this research, we focus on detecting NF patients in order to find timely treatment before symptoms become severe. Therefore, we focus on the desired model with a high recall or true positive rate percentage, which is the decision tree (DT) model.

Understanding which features are most important allows us to interpret the model’s decisions and improve its reliability. Figure 1 below illustrates the relative importance of the six key features used in the model, demonstrating how much each contributes to the overall prediction.

Figure 1. The bar chart highlights that hemoglobin (HB) and white blood cell counts (WBC) are the most critical factors in the model’s predictions, platelet count (PT), C-reactive protein (CRP), and creatinine (CT) also play important roles, while sodium levels (NA) have the least influence among the features shown. This suggests that these features are crucial for the model to accurately identify or predict the medical condition in question, with HB and WBC being particularly significant. Platelet count is a key variable for distinguishing NF patients from those with cellulitis, ranking as the third most important feature. Furthermore, the findings of Chen et al. [10] indicate that platelet count is a valuable and cost-effective prognostic indicator for NF patients. Therefore, we incorporate platelet count to refine the LRINEC score for the improved detection of NF patients, which will be discussed in the next section.

3.2. Modified Laboratory Risk Indicator for Necrotizing Fasciitis (MLRINEC) Score

In this research, we used six laboratory values to adjust the LRINEC score. Five of these values—creatinine (CT), hemoglobin (HB), sodium (NA), white blood cells (WBCs), and C-reactive protein (CRP)—were applied as originally described by Wong et al. [3]. However, we substituted platelet count (PT) for glucose in our analysis. Consequently, to determine the PT score, we will consider the following statistical values and the graph in Figure 2.

Figure 2 displays a box plot comparing platelet counts (PT) between two patient groups: those with cellulitis and those with necrotizing fasciitis (NF). The plot shows that patients with NF have a higher median platelet count and greater variability compared to those with cellulitis. Outliers in the cellulitis group indicate that some patients had significantly elevated platelet counts, which may be clinically significant. The red lines mark the normal platelet count range, typically between 150,000 and 440,000 cells/mm³. The NF group consistently shows higher mean, median, and upper percentile values, reflecting the greater variability observed in the box plot. However, some patients in both groups have similar platelet counts. To categorize the PT scores, the following modified scores are applied: PT less than mean minus standard deviation is defined as 0, PT between mean minus and mean plus standard deviation is defined as 1, and PT greater than mean plus standard deviation is defined as 2.

Below is

Table 5. The modified laboratory risk indicator for NF (LRINEC) score.

Variable	Result	Score
Creatinine level, mg/dL	≤1.6	0
	>1.6	2
Hemoglobin, g/dL	>13.5	0
	11–13.5	1
	<11	2
Platelet: PT (cells/mm3)	<189,870	0
	189,870–525,010	1
	>525,010	2
Sodium Level (mmol/L)	≥135	0
	<135	2
WBC count (cell/mm3)	<15,000	0
	15,000–25,000	1
	>25,000	2
C-reactive protein (CRP), (mg/L)	<150 (Negative)	0
	≥150 (Positive)	4

The maximal scoring is 14, score ≤ 5 = <50% risk (low); 6–7 = intermediate risk; ≥8 => 75% risk (high).

which provides the modified Laboratory Risk Indicator for Necrotizing Fasciitis (LRINEC) score. This scoring system is used to assess the risk of necrotizing fasciitis, a severe soft tissue infection. The score is calculated based on several laboratory parameters, with higher scores indicating a greater risk of the condition.

Table 5 provides a structured method to evaluate the risk of necrotizing fasciitis in patients by assigning scores to specific laboratory results. The score range for determining the risk level was based on the original by Wong et al. [4]. The total score helps clinicians decide the urgency and nature of further diagnostic and treatment steps, ensuring timely and appropriate care for patients at risk of this severe condition.

Table 6. Number and percentage of classification of necrotizing fasciitis and cellulitis patients using risk level from modified score.

Risk Levels from Modified Score	Necrotizing Fasciitis: NF	Cellulitis	Total
Low risk (Score ≤ 5)	46	55	101
	45.5%	54.5%	100.0%
Intermediate risk (score 6–7)	22	12	34
	64.7%	35.3%	100.0%
High risk (score ≥ 8)	6	3	9
	66.7%	33.3%	100.0%
Total	74	70	144
	51.40%	48.60%	100.0%

provides a breakdown of the number and percentage of patients diagnosed with necrotizing fasciitis (NF) and cellulitis, categorized according to their risk level based on the modified Laboratory Risk Indicator for Necrotizing Fasciitis (MLRINEC) score. The risk levels are divided into three categories: Low risk, Intermediate risk, and High risk. The key takeaway is that as the risk level increases, the proportion of patients diagnosed with necrotizing fasciitis also increases, suggesting that the MLRINEC score is effective in distinguishing between these two conditions. This is especially important for clinical decision making, where the early identification and treatment of necrotizing fasciitis are critical due to its severity.

4. Discussion

The study demonstrates that combining traditional risk-scoring methods with machine learning models can enhance the diagnostic accuracy for NF. The choice of the most appropriate model should be guided by the clinical context—whether the goal is to maximize the detection of NF (favoring recall) or to minimize false positives (favoring specificity). These findings have significant clinical implications. The use of machine learning models, particularly the decision tree model, can aid clinicians in making timely and accurate diagnoses of NF, reducing the risk of severe complications.

The modified LRINEC score that substitutes platelet count for glucose levels reflects the importance of this parameter in the diagnosis of NF. The score assigns points based on the ranges of parameters, with higher scores indicating a greater risk of NF. The inclusion of platelet count as a key variable is particularly noteworthy. The study by Chen et al. [10] highlighted the prognostic value of platelet count in NF patients, and the current study reinforces this by ranking platelet count as the third most important feature. This suggests that platelet count should be considered in the diagnostic process, potentially refining the existing diagnostic tools like the LRINEC score. This modification addresses the data limitations in our hospital, where the glucose and fibrinogen levels needed for calculating the original LRINEC or the MLRINEC score suggested by Thomas et al. (2015) [4] are unavailable. These gaps are due to constraints in the laboratory tools used. By refining the score, this study enhances its clinical utility, potentially enabling earlier and more accurate diagnoses of NF.

Future research could explore the integration of these models into clinical decision support systems, allowing for real-time risk assessment and diagnosis in emergency and inpatient settings. Additionally, further studies could validate the modified LRINEC score in larger, more diverse populations to confirm its generalizability and effectiveness across different healthcare settings.

5. Conclusions

Machine learning techniques, particularly the decision tree (DT) classifier, are effective in differentiating necrotizing fasciitis (NF) from other soft tissue infections like cellulitis. The DT model exhibited the highest accuracy in identifying NF cases with a recall rate of 100%, making it the most suitable model for early detection and treatment. While Logistic Regression (LR) offered high precision and specificity, it had moderate recall. Random Forest (RF) showed balanced performance but struggled with specificity. Overall, the choice of the best model depends on the application’s specific needs, whether prioritizing sensitivity to detecting positive cases (recall) or accuracy in identifying negative cases (specificity). As a result of this research, we recommend using the decision tree model because it gives the highest recall or true positive value of 100 percent. The positive group is the NF patient group. If we can use the model to accurately predict this group, it will lead to receiving urgent treatment and can reduce the mortality rate for NF patients. Feature importance analysis revealed that hemoglobin (HB) and white blood cell counts (WBC) were the most significant predictors in distinguishing NF from cellulitis. The platelet count (PT), C-reactive protein (CRP), and creatinine (CT) also played important roles, while sodium levels (NA) were the least influential.

Moreover, the research modifies the Laboratory Risk Indicator for Necrotizing Fasciitis (LRINEC) score by substituting platelet counts for glucose levels in the original LRINEC score from [3], thereby improving the tool’s predictive capability. Combining traditional risk scoring methods with machine learning models offers a more accurate assessment of NF risk, potentially enhancing patient outcomes by enabling earlier and more precise intervention. The inclusion of platelet count as a key variable is especially noteworthy. The findings suggest that incorporating advanced analytical techniques and refining the existing diagnostic tools can significantly enhance clinical decision making and patient care.