This was a retrospective cohort study. A prognostic prediction analysis and clinical score development study was conducted in accordance with the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement (19). Ten outcome events per predictor component were used to determine the sample size (the EPV method) (20). Based on other similar scoring systems that collectively draw upon 10 variables, a total of 100 amputations NF were needed.

From January 2009 to December 2012, the medical records of patients with surgically-confirmed NF were obtained from three general hospitals in Northern Thailand. The hospitals were Chiangrai Prachanukroh Hospital (a 600-bed tertiary-care medical center), Kamphaeng Phet Hospital (a 330-bed public hospital), and Phayao Hospital (a 400-bed secondary-care medical center).

In total, 1,506 patients with NF were enrolled. They were divided into two groups using random sampling by computer generation (4:1), Stata command “insample.” The derivation cohort (n = 1,193) is a sample to develop ANF scoring system, and a validation cohort (n = 313) is to confirm accuracy and appropriateness of this scoring system. In our study, an NF with amputation was found in 127 patients (8.4%). This detail is shown in Table 1.

Necrotizing fasciitis was defined by the presence of extensive necrosis affecting at least the fascia and subcutaneous tissue. Patients diagnosed with NF according to the standard definition were eligible for inclusion. Diagnoses were made by surgeons who strictly observed the guidelines for skin and soft tissue infections produced by the Infectious Diseases Society of America (21).

The stages of wound appearance were defined as a clinical stage of NF. Cutaneous manifestations of NF present when the disease progresses through stages 1, 2, and 3 (early, intermediate, and late stages). Wound appearance stage 1 (early): the disease was clinically presenting with swelling and erythema; stage 2 (intermediate): belb formation is an important diagnostic clue of NF. When present, it signals the onset of critical skin ischemia; stage 3 (late): indicates the development of tissue necrosis and skin gangrene, which are severe signs of NF infection (21, 22).

Severe pain was defined based on the patient complaint by use of a pain-assessment tool (numerical rating scale 7–10) according to the pain management guideline 2017 (23, 24).

The protocol for this study was approved by the Ethics Committee of Chiangrai Prachanukroh Hospital and the Faculty of Medicine, Chiang Mai University, Chiang Mai, Thailand (032/2013; research ID: 1461; study code: COM-13-1461-EX).

Descriptive statistics were used to summarize patient demographic and clinical characteristics. Continuous data were compared using Student's t-test, and those results are shown as mean ± standard deviation (SD). Categorical data were compared using chi-square test or Fisher's exact test, and those results are given as number and percentage. All p-values are two-tailed and were considered statistically significant at a p-value < 0.05.

Variables that could affect the outcomes were established and incorporated into the development model. The predictive variables were selected by factors reported from previous studies to ensure a clinical significance (25–27). Clinically meaningful and interpretable cut-off points were used to transform continuous variables into categorical variables. In addition, we also selected the parameters from statistical analysis that P-values < 0.05 were considered statistically significant to include in running the model. Amputation and no amputation variables were evaluated using univariable logistic regression, and those results are summarized as odds ratio (OR) and 95% confidence interval (CI). Variables found to be statistically significant and other variables of interest were then entered into a multivariable logistic regression model to identify factors independently associated with amputation in NF. Item scores were calculated using the weighted coefficients of the significant variables from multivariable analysis. By dividing each regression coefficient by the smallest coefficient of the model and rounding to the nearest integer, the score was determined (28–30). To determine the power of the model to discriminate between amputation and no amputation, a receiver operating characteristic (ROC) curve was plotted, and the area under the ROC curve (AuROC) was also calculated. A Hosmer–Lemeshow goodness of fit test was also performed (31). Cut-off scores were determined to classify the patients with NF as low-risk, moderate-risk, or high risk for death. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NVP), positive likelihood ratio (LR+), and negative likelihood ratio (LR–) were calculated to evaluate the diagnostic performance of the model (32).

The validation cohort (n = 313) was used to verify the scoring system. The efficiency and accuracy of the model were assessed using a ROC curve. The same diagnostic performance parameters that were calculated and assessed in the derivation cohort were calculated and assessed in the validation cohort.

A total of 1,506 patients with NF were enrolled. Their ages ranged from 2 to 95 years, 43.69% were female, 85.29% had a body mass index (BMI) ≥18.50, and 25.70% had diabetes mellitus. A swelling wound (wound appearance stage 1) was presented in 94.82% of patients, and the most common wound site was the lower limbs (77.04%). An amputation with NF found 127 patients (8.4%) and non-amputation with NF found 1,379 patients (91.6%). Sites of amputation were finger/toes (44.1%), above knee (22.8%), below knee (20.5%), forefoot (8.7%), hand and forearm (3.9%). Clinical and demographic characteristics of patients with NF among all patients and comparison between the derivation cohort and the validation cohort are shown in Table 1.

The overall amputation rate in this study was 8.43% (127 of 1,506 patients). Univariable analysis for risk factors significantly associated with amputation in patients with NF in the derivation cohort is shown in Table 2. Factors found to be significantly associated with amputation were female gender, age, education, heart disease, hypertension, erythematous wound, bleb wound, white blood cell count (WBC), percentage of polymorphonuclear cell or neutrophil, serum creatinine, serum bicarbonate, pulse rate, systolic blood pressure, diastolic blood pressure, severe sepsis, and length of hospital stay.

The variables identified in univariable analysis were entered into subsequent multivariable analysis to develop the scoring system for amputation in patients with NF. Using backward stepwise logistic regression, the multivariable model included four variables. The weighted coefficient was used to calculate the point value for each factor, which was then rounded to the nearest whole number (Table 3). The total prediction score comprises the summation of the following four scores: female gender (yes = 1, no = 0); diabetes mellitus (yes = 3, no = 0); wound appearance stage 3 (yes = 3, no = 0); and creatinine ≥1.6 mg/dL (yes = 2, no = 0). The highest possible score is 9, and a higher score indicates a higher risk for amputation.

To categorize patients into three risk categories, cut-off scores of 4 and 7 were selected. Patients with a total score of ≥7 were considered to be at high risk. Amputation among those patients was found to be predicted with moderate accuracy (28/78 cases; PPV 35.89%). Patients with a total score of ≤4 were considered to be at low risk. No amputation was correctly predicted in 95.99% of cases (863/899) in this group. No amputation group could be excluded with moderate accuracy (NPV 21.43%). The corrected prediction of absence or presence of amputation was (863 + 28)/(899 + 78) = 91.19%, whereas the incorrect prediction rate was (36 + 50)/(899 + 78) = 8.80% (Table 4). With this scoring system and using two cut-off points, the risk score discriminated between patients with NF with a risk for amputation and those without a risk for amputation with high validity (AuROC 76.68%) (Figure 1). The predictive model was also shown to be well-calibrated (Hosmer-Lemeshow χ² = 2.59; p = 0.8586).

The ROC curves for the derivation and validation cohorts showed similar results (AuROC 76.68 and 76.82%, respectively; Figure 1). Amputation in the high-risk group was accurately predicted in 36.84% of cases, and the presence of amputation was predicted with high moderate accuracy (PPV 36.84%). In the low-risk group, no amputation was predicted in 94.51% (224/237) of the cases, with the absence of amputation being predicted with moderate accuracy (NPV 34.75%). The accurate prediction of amputation or no amputation was (224 + 7)/(237 + 19) = 90.23%, whereas the incorrect prediction rate was (13 + 12)/(237 + 19) = 9.76% (Table 5).

The probability of amputation in patients with NF stratified by risk score and compared between the derivation and validation cohorts is shown in Figure 2. Patients were classified into three groups based on the calculated cut-off values, as follows: low-risk (score ≤4), moderate-risk (score of 5–6), or high-risk (score ≥7). All comparisons among the three risk groups were statistically significant with p-values <0.001 for all comparisons in both the cohorts.

The strengths of this study should be emphasized. First, this study included a large sample size of patients with NF to evaluate amputation outcome. We also included patients from three large hospitals in Thailand, which suggests that our results can be generalized to other regions of Thailand, and to other low to middle income Asian countries. Second, the ANF risk scoring system was developed in accordance with the stringent criteria set forth in the TRIPOD statement (19). Third, the developed scoring system comprises only four variables, and all of them are easy to obtain and quantify. Interestingly, the predictors that were identified are different types of parameters. ANF scoring includes patient demographic data (being female), patient comorbidity data (diabetes mellitus), patient clinical characteristic data (wound appearance stage 3, presenting with skin necrosis and gangrene), and patient routine laboratory data (serum creatinine). Fourth, the ANF risk scoring system was validated using different patient data sets. In both the derivation and validation cohorts, the ANF score demonstrated good prediction ability and satisfactory diagnostic results. Fifth and last, since serum creatinine is the only laboratory test that is needed, this model is very affordable. The cost of this investigation is only 60 Thai baht (THB) (2 USD), and serum creatinine is a routine laboratory test that is routinely used to track patient health status while they are in the hospital.

The most notable limitation of this study is it retrospective design, which rendered it vulnerable to missing or incomplete data. For example, the LRINEC scoring system could not be used in this study because some important laboratory investigations, such as CRP level, were not performed in all patients.

The data on the amputations index-date, time, and duration of index-operations were not collected because of the retrospective analysis from January 2009 to December 2012. However, since the reason of performing an amputation is to control the infection indicated by several outcomes, e.g., reducing length-of-stay (LOS) or mortality rate, we found that LOS in a hospital of non-amputated is significantly lower (11 days, SD = 11.59) than those in amputated groups (16 days, SD = 10.79), p < 0.001. However, there was no difference in mortality between amputations and no amputations (20.20 vs. 19.12 %, Odds ratio 1.07: 95% Cl 0.61–1.81). The data on the amputations index date, time, and duration of index operations were not collected. However, we have the data of length of stay in a hospital of non-amputated, which was 11 days (SD = 11.59) and in amputated it was 16 days (SD = 10.79), with a P-value < 0.001. Amputation was performed to control the infection. However, there was no difference in mortality between amputations and no amputations. Our data showed that the mortality of amputated was 20.20% and non-amputated was 19.12% (Odds ratio 1.07; 95% Cl 0.61–1.81). This was similar to previous studies (1, 26) which reported that amputation did not show the reduction of mortality, but patients who underwent amputation had to undergo fewer operations to control the infection and to achieve wound coverage. Kaplan–Meier survivorship analysis revealed that the survival rate decreased with delayed surgery and prolonged symptoms. Survival sharply declined with a delay in surgery of more than 24 h.

Despite this limitation, the ANF risk scoring system can be used in all clinical settings because of its low-cost and because it is simple to implement and use. The advantage of the ANF score is that it can help to detect early the amputation risk in patients with NF. Using this system, disease progression can be stopped or reduced, potential complications of NF can be monitored, and the risk of amputation can be greatly reduced or eliminated.

The developed ANF risk scoring system, which includes four easy to obtain predictors, provides physicians with prediction indices for amputation in patients with NF. This model will assist clinicians with surgical decision-making in this time-sensitive clinical setting.