Necrotizing fasciitis (NF) is a type of necrotizing soft tissue infection characterized by tissue destruction, systemic signs of toxicity, and high mortality rates. It may involve the extremities, perineum, and head and neck region.¹ Diagnosis can be challenging as it may be confused with other diseases.² NF of the perineum is called Fournier’s gangrene (FG).^1,3 There are several risk factors for this life-threatening necrotising soft tissue infection, including trauma, obesity, immunosuppression, malignancy, alcoholism, and diabetes mellitus (DM).¹

Although DM is one of the important predisposing factors for NF,^1,3,4 it has not been found to be consistently associated with mortality in the literature. In addition, there are a few case reports of sodium-glucose co-transporter 2 (SLGT2) inhibitors (a new oral hypoglycemic agent) inducing FG (in males more than females).^4-6 Based on these case reports, the US Food and Drug Administration (FDA) has warned of the occurrence of this rare, serious infection (i.e., FG) with the use of SGLT2 inhibitors. FDA data have shown that the average time to onset of FG after SGLT2 inhibitor initiation is 9.2 months (range, 7 days - 25 months).⁵ There are also studies in literature about the association of DM with FG.

Although not fully validated, there is a scoring system to diagnose NF (e.g., the LRINEC, or Laboratory Risk Indicator for Necrotizing Fasciitis Score).⁷ There is also a scoring system for predicting mortality in FG [the Fournier's Gangrene Severity Index (FGSI) score]. Some parameters of this scoring system require input from the managing physician (such as heart rate, respiratory rate).^7,8 The LRINEC score is used for the diagnosis of NF.⁹It has been suggested that DM, causative pathogens, and other patient-related parameters could also predict outcomes in NF.^3,10 Determining early admission and early follow-up routine patient data that could affect treatment outcomes, and the development of a scoring system, if possible, from these associated factors is of paramount importance to physicians and healthcare providers, especially for a somewhat rare and life-threatening disease such as NF.

Therefore, the aim of this retrospective study was to determine the predictive probability of hospital mortality from the determined associated risk factors for this disease.

This was a single-center, retrospective study. Data were recorded for surgery-proven NF cases admitted to Bakirkoy surgical department between 1 January 2016 and 31 December 2018. Exclusion criteria were as follows: <18 years old, incomplete recording of patient data (to minimize this, records of the last 3 years were included), and referral to another hospital before completing patient management. A total of 110 patients were included in this study. The records of 4 patients were excluded for the following reasons: two male patients were excluded because of incomplete data recording, one male patient was excluded because of referral to another hospital on the second postoperative day (lack of beds), and one female patient self-discharged before completing the management. The inclusion and exclusion of cases was decided by a panel of three researchers (G.S. Erdal, S. Ferahman, and N. Kocamaz). A record was made of demographic data (age, gender), presence or absence of DM (including use of an SGLT2 inhibitor), and complete blood count (CBC) parameters, serum creatinine, and C-reactive protein (CRP) on first presentation at hospital. The wound and operative materials culture (referred to hereafter as "wound culture”) results were recorded and analyzed.

Approval for this retrospective study was granted by the Ethics Committee of Bakirkoy Dr. Sadi Konuk Training and Research Hospital (study protocol code: 2018-444). Hereafter, this hospital or center will be abbreviated as "Bakirkoy.” Written permission to collect the data was obtained from the administrative authority of Bakirkoy. Written or verbal consent was obtained from the patients, or their relatives, where applicable.

Statistical analysis

Data obtained in the study were analyzed statistically using R 3.5.3 Statistical Software (https://cran.r-project.org/). Baseline characteristics of the patients that were normally distributed were presented as mean, standard deviation (SD), and minimum/maximum values, and variables that were not normally distributed as median, interquartile range (IQR), and minimum/maximum values. Conformity of the data to normal distribution was assessed visually with histograms and with the application of the Shapiro–Wilk test.

The univariate logistic regression model was fit to determine the clinical characteristics and risk factors for mortality. The multivariable logistic regression model was fit using stepwise variable selection based on the minimum Akaikeinformation criterion (AIC) value.¹¹ Best subsets variable selection was also applied using the "bestglm” package based on minimum BICq, which gave the same model.¹² Models were fitted using logistic regression of the formula, Probability = e^A(x)/1+e^A(x), where A(x) = constant + sum of (variables * associated regression coefficients).

Model performance was evaluated through the area under the receiver operating characteristic (ROC) curve (AUC), coefficient of discrimination (Tjur’s D) and the Hosmer-Lemeshow (HL) goodness-of-fit test.McFadden, Cox, and Snell and Nagelkerke’s pseudo-R-squared values for both the full and the reduced model were calculated using the "nagelkerke” function in the "rcompanion” package.

There was no evidence of multicollinearity between covariates in the logistic regression models (variance inflation factor <1.8 for all independent variables). The Cook’s distance and standardized residuals were calculated, and no potential outliers were detected.

ROC analyses were conducted to determine the cut-off value for significant risk factors of mortality. The cut-off points were selected based on maximizing the Youden’s index. The pROC, Epi, and Optimal Cutpoints packages were used to determine the optimal cut-off points.^13,14 A nomogram was constructed from a multivariable logistic regression prediction model with the following covariates: age, sex, DM, creatinine, PLT, presence of Acinetobacter or Klebsiella spp. in wound culture.¹⁵

All p values were two-sided, and p <0.05 was considered statistically significant.

A power analysis was conducted. The minimum required sample size for an area under the ROC curve of 0.70 was 100 patients (assuming 95% power and alpha 0.05). The power analysis was conducted using the "power.roc.test” function in the pROC package.¹³

Evaluation was made of the data of a total of 106 NF patients. The patients were equally distributed with respect to gender, and the mean age of the sample was 58 ±16 years (range: 19-87 years). Death occurred in 30% of the patients. The demographic and laboratory test results of the patients on admission are listed in Table 1.

Risk factors associated with mortality are listed in Table 2. In univariable logistic regression, increases in age, hospital stay, creatinine, and the neutrophil-lymphocyte ratio(NLR) were associated with an increased risk of mortality and decreases in lymphocyte, absolute lymphocyte, hemoglobin (Hgb), hematocrit % (Hct), and PLT were also associated with an increased risk of mortality in NF patients.

 Table 2. The estimated coefficients in the univariate logistic regression (left) and reduced multivariable logistic regression (right) with descriptors and clinical demographic data where the dependent variable is exitus status

The presence of DM and the presence of Acinetobacter or Klebsiella spp. in wound cultures were associated with an increased risk of mortality. It should be noted that the ORs obtained with univariate analyses are unadjusted. The AUC values and cut-off points to predict mortality for NF (unadjusted) are shown in Table 3.

The multivariable model was obtained using stepwise regression based on Akaike information criteria (AIC) minimization, including age, DM, creatinine, PLT, and the presence of Acinetobacter or Klebsiella spp. in wound culture, which were all determined to be significant risk factors (p<0.05) (Table 4).

With control for age, serum creatinine, PLT, and the presence of Acinetobacter or Klebsiella spp. in wound cultures, patients with DM were found to have a higher risk of mortality than patients without DM (OR=3.9, 95% CI: 1.22-13.70,p=0.026). Increases in serum creatinine and decreases in serum PLT increased the risk of mortality. The presence of Acinetobacter or Klebsiella spp. in wound cultures was found to be highly associated with the risk of mortality, with very high odds ratios of 30 and 98, respectively, controlling for other variables in the reduced model.

As a widely used measure of the discriminatory power of a logistic model, the area under the ROC curve was 0.904 (95% CI: 0.848-0.960) (Figure 1). The coefficient of discrimination (Tjur’s D) was 0.470. The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) values of the final model were 71.9%, 93.2%, 82.1%, and 88.5%, respectively. The goodness of fit of the reduced model was good as measured using the Hosmer-Lemeshow test (10.22 based on 8 degrees of freedom, p=0.250). The reduced model in Table 2 has a good predictive power of mortality and can be calculated to find the risk of death for a patient given some variables. The optimal model to predict mortality for NF is equal to e^A(x)/1+e^A(x)(where A(x)=- 6.458 + 0.100Age + 1.361DM(=1) + 0.848 creatinine - 0.008 PLT + 4.586 wound: Klebsiella spp.(=1) + 3.413 wound: Acinetobacter spp. (=1)). The McFadden, Cox, and Snell and Nagelkerke pseudo-R-squared values for this optimal model were 0.498, 0.456, and 0.646, respectively, which are larger than the pseudo-R-squared values of the full model.

The cut-off for the fitted probabilities of the optimal model was selected as 0.5 based on 0.877 model accuracy (95% CI: 0.799-0.933), 0.719 sensitivity, 0.946 specificity, 0.852 positive predictive value, and 0.886 negative predictive value (Figure 2). The ROC curve of the optimal model and its area under the ROC curve is shown in Figure 2.

A nomogram was drawn for the reduced model. The nomogram in Figure 3 is the graphical representation of the logistic regression formula in Table 2. A vertical line drawn from the scale value of each predictor up to the points provides the numerical score for that predictor. The total of the 7 predictors yields the total points, which can be scaled to the final output probability of mortality. For example, for a patient who is 65 years old, DM (present), creatinine = 1.2, PLT = 244000, Wound: Acinetobacter spp. (absent), Wound: Klebsiella spp. (absent), the points are 62.5, 17.5, 12.5, 72.5, and 0,0. The total score is then 165, and the corresponding risk of death according to the nomogram is 0.49 as given in Table 4.

There were 16 DM cases in the survivor group and 15 DM cases in the non-survivor group. None of these patients were using an SGLT2 inhibitor, but 4 of the survivors and 13 of the non-survivors were using insulin treatment alone. One of the 2 remaining diabetic non-survivors was using metformin with insulin and the other diabetic non-survivor was using only gliclazide tablets. The remaining 12 diabetic patients of the survivor group were using oral antidiabetics (4 patients using metformin alone, 3 patients using gliclazide alone, 1 patient using sitagliptin alone, 1 patient using metformin plus gliclazide, 1 patient using metformin plus linagliptin, 1 patient using metformin plus sitagliptin, and 1 patient using metformin plus Exenatide).

The study center (Bakirkoy) is one of the largest tertiary level hospitals in Istanbul, Turkey. According to the hospital statistics, approximately 42,000 surgical patients present at the Emergency Department per year (excluding elective outpatient applications). In spite of this, the 3-year sample size of NF did not exceed 110 patients. This may be somewhat explained by the rarity of this disease. NF is a life-threatening disease that needs early, prompt management. There are studies in literature related to the prediction of severity and/or mortality.^10,16,17 These studies have used clinical and laboratory data to predict prognosis. Hansen et al.¹⁶ found plasma pentraxin-3 (PTX3) to be a good biomarker to predict mortality in NF. Baseline PTX3 levels above the median were found to be associated with mortality (p=0.009). The finding of that study is interesting as the PTX3 normal range is not known in the healthy population in daily practice, and it could be affected by other comorbidities, such as kidney failure.¹⁸ The current study used easily obtainable patient data to predict treatment outcomes (Tables 2 and 3). In FG, which is a variant of NF, the Fournier's Gangrene Severity Index (FGSI) has been found to be useful in predicting the outcomes. The FGSI system consists of clinical and laboratory parameters.⁷ Developing a scoring system for rare diseases and/or with somewhat limited sample sizes is difficult, but not impossible.^19,20 Therefore, in this study we have developed a final (reduced) model (suggested by the authors Hursitoglu & Akdeniz) and a nomogram of the logistic regression model of the probability of mortality (see Table 4 and Figures 2 and 3). The area under the ROC curve for this ‘’Bakirkoy’’ model is 0.904 (95% CI: 0.848-0.960). The goodness of fit of this model was good as measured using the Hosmer-Lemeshow test (see above). The determined predictors of mortality for NF in this study were age, serum creatinine, PLT, DM, and the presence of Klabsiella spp. or Acinetobacter spp. in wound culture. These early admission parameters could help the managing physician teams and/or healthcare providers to make preliminary predictions of the mortality risk of NF cases on admission (by ignoring wound culture results) and/or evaluating their center outcomes retrospectively to help them to revise management approaches in the future. However, it should be mentioned that this pilot model for NF mortality prediction was based on the patient data of a single center and therefore, it needs to be replicated and validated by future studies.

In NF (including FG), DM is an important precipitating factor.^3,7,10,17,21 In addition, SLGT2-inhibitor induced FG cases have been reported.^4,5 Based on these case reports, the FDA has warned of the occurrence of this rare, serious infection (FG) with the use of SGLT2 inhibitors.⁶ In Turkey, two SGLT2 inhibitors have been in use since September 2016: first dapaglifozin (Forziga^®) and later empagliflozin (Jardiance^®) entered the Turkish antidiabetic drug market. The prescription rate of this drug by this study center’s physicians is about 2-3%. In the current study, there was a significant association between NF (including FG) mortality and the presence of DM. Interestingly, none of the NF and/or FG patients were using an SGLT2 inhibitor at the time of having this infectious disease. As mentioned above, one patient was using the dipeptidyl peptidase 4 (DPP-4) inhibitor linagliptin, and another was using the glucagon-like peptide-1 (GLP-1) inhibitor exenatide. Both of these patients were in the survivor group. The other patients were using other antidiabetics (insulins, sulphonylureas, etc.). The last published DECLARE–TIMI 58study with dapaglifozin also showed no relationship between this drug and FG, although the FG rate was somewhat higher in the placebo group.²² Therefore, it appears that it is mostly DM, and not SGLT2 inhibitors, which is a precipitating factor of FG. The last published consensus report of the American Diabetes Association /European Association for the study of Diabetes (ADA/EASD) recommended using an SGLT2 inhibitor as a second line or first line, if metformin is contraindicated, of defense in type 2 DM.²³ Nevertheless, recently published studies still emphasize the above-mentioned FDA warning.²⁴ The FDA will most likely revise this warning about the use of SGLT2 inhibitors with the emergence of additional studies.

Limitations

This was a retrospective study, and another possible limiting aspect of this study was that the data used were from a single center in a single country. However, the associated risk factors of mortality determined in this study have also been reported in some similar previous studies from other countries (e.g., age, DM, serum creatinine level, etc.).^3,10,25,26 A study by Audureau et al. showed a difference in the mortality rate of NF between teaching and non-teaching public hospitals of France.²⁷ Whether the condition is the same in Turkey is not known. Therefore, taking the above-mentioned points into consideration, there is a need for further detailed studies on this issue, possibly using the model suggested in this paper.

Although SGLT2 inhibitors have been thought to be responsible for the pathogenesis of FG, surprisingly, none of the current study diabetic FG or NF patients were using this drug. Another interesting finding of this study was the determination of a significant association of obtainable patient data with the mortality of NF. Using these simple parameters, we have developed a pilot model (Bakirkoy model) that shows good performance in predicting mortality risk. However, further studies are still needed to confirm this.

1. Stevens DL, Baddour LM. Necrotizing soft tissue infections. 2018. In: Melin JA (Ed.), UpToDate. Accessed on: 06 December 2018. Available at: https://www.uptodate.com/contents/necrotizing-soft-tissue-infections.

2. Sanchez IM, Lowenstein S, Johnson KA, et al. Clinical features of neutrophilic dermatosis variants resembling necrotizing fasciitis. JAMA Dermatol. 2019;155:79-84.

https://doi.org/10.1001/jamadermatol.2018.3890

3. Cheng NC, Tai HC, Chang SC, Chang CH, Lai HS. Necrotizing fasciitis in patients with diabetes mellitus: Clinical characteristics and risk factors for mortality. BMC Infect Dis. 2015;15:417.

https://doi.org/10.1186/s12879-015-1144-0

4. Kumar S, Costello AJ, Colman PG. Fournier's gangrene in a man on empagliflozin for treatment of Type 2 diabetes. Diabet Med. 2017;34:1646-8.

https://doi.org/10.1111/dme.13508

5. Chi W, Lim-Tio S. Fournier's syndrome: a life-threatening complication of SGLT2 inhibition in poorly controlled diabetes mellitus. Australian Diabetes Society (ADS) and Australian Diabetes Educators Association (ADEA) 2016 Annual Scientific Meeting; 24-26 August 2016; Gold Coast, Australia. Abstract number 265.

6. US Food and Drug Administration. FDA warns about rare occurrences of a serious infection of the genital area with SGLT2 inhibitors for diabetes. Accessed on: 10 December 2018. Available at: https://www.fda.gov/downloads/Drugs/DrugSafety/UCM618466.pdf

7. Verma S, Sayana A, Kala S, Rai S. Evaluation of the utility of the Fournier's Gangrene Severity Index in the management of Fournier's gangrene in North India: a multicentre retrospective study. J Cutan Aesthet Surg. 2012;5:273-6.

https://doi.org/10.4103/0974-2077.104916

8. Pehlivanlı F, Aydin O. Factors affecting mortality in Fournier gangrene: A single center experience. Surg Infect (Larchmt). 2019;20:78-82.

https://doi.org/10.1089/sur.2018.208

9. Bechar J, Sepehripour S, Hardwicke J, Filobbos G. Laboratory risk indicator for necrotising fasciitis (LRINEC) score for the assessment of early necrotising fasciitis: a systematic review of the literature. Ann R Coll Surg Engl. 2017;99:341-6.

https://doi.org/10.1308/rcsann.2017.0053

10. Khamnuan P, Chongruksut W, Jearwattanakanok K, Patumanond J, Yodluangfun S, Tantraworasin A. Necrotizing fasciitis: risk factors of mortality. Risk Manag Healthc Policy. 2015;16:1-7.

https://doi.org/10.2147/RMHP.S77691

11. Hosmer DW Jr., Lemeshow S, Sturdivant RX. Applied logistic regression, third edition. New Jersey: John Wiley & Sons, Inc; 2013.

https://doi.org/10.1002/9781118548387

12. McLeod AI, Xu C. Bestglm: Best subset GLM. R package, version 0.33. Accessed on: 16 March 2019. Available at: http://CRAN.R-project.org/package=bestglm

13. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77.

https://doi.org/10.1186/1471-2105-12-77

14. Lopez-Raton M, Rodriguez-Alvarez MX, Cadarso-Suárez C, Gude-Sampedro, F. OptimalCutpoints: An R package for selecting optimal cutpoints in diagnostic tests. J Stat Softw. 2014;61:1-36.

https://doi.org/10.18637/jss.v061.i08

15. Zhang Z, Kattan MW. Drawing nomograms with R: Applications to categorical outcome and survival data. Ann Transl Med. 2017;5:211.

https://doi.org/10.21037/atm.2017.04.01

16. Hansen MB, Rasmussen LS, Garred P, Bidstrup D, Madsen MB, Hyldegaard O. Pentraxin-3 as a marker of disease severity and risk of death in patients with necrotizing soft tissue infections: a nationwide, prospective, observational study. Crit Care. 2016:20:40.

https://doi.org/10.1186/s13054-016-1210-z

17. Ray-Zack MD, Hernandez MC, Younis M, et al. Validation of the AAST EGS grade for skin and soft tissue infection. J Trauma Acute Care Surg. 2018;84:939-45.

18. Honore PM, Spapen HD. Pentraxin-3 to better delineate necrotizing soft tissue infection: not really! Crit Care. 2016;20:173.

https://doi.org/10.1186/s13054-016-1319-0

19. Zhang Z, Zhang H, Khanal MK. Development of scoring system for risk stratification in clinical medicine: a step-by-step tutorial. Ann Transl Med. 2017;5:436.

https://doi.org/10.21037/atm.2017.08.22

20. Soto A, Solari L, Agapito J, et al. Development of a clinical scoring system for the diagnosis of smear-negative pulmonary tuberculosis. Braz J Infect Dis. 2008;12:128-32.

https://doi.org/10.1590/S1413-86702008000200006

21. Tan JH, Koh BT, Hong CC, et al. A comparison of necrotising fasciitis in diabetics and non-diabetics: a review of 127 patients. Bone Joint J. 2016;98-B:1563-8.

https://doi.org/10.1302/0301-620X.98B11.37526

22. Wiviott SD, Raz I, Bonaca MP, et al. Dapagliflozin and cardiovascular outcomes in type 2 diabetes. N Engl J Med. 2019;380:347-7.

https://doi.org/10.1056/NEJMoa1812389

23. Davies MJ, D'Alessio DA, Fradkin J, et al. Management of hyperglycemia in type 2 diabetes, 2018. A Consensus Report by the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD). Diabetes Care. 2018;41:2669-701.

https://doi.org/10.2337/dci18-0033

24. Bersoff-Matcha SJ, Chamberlain C, Cao C, Kortepeter C, Chong WH. Fournier gangrene associated with sodium-glucose cotransporter-2 inhibitors: a review of spontaneous postmarketing cases. Ann Intern Med. 2019;170:764-9.

https://doi.org/10.7326/M19-0085

25. Melis A, Riu F, Kihlgren C, et al. Medical-surgical management and clinical outcome in cervical abscesses. J Infect Dev Ctries. 2020;14:527-31.

https://doi.org/10.3855/jidc.12191

26. Jabbour G, El-Menyar A, Peralta R, et al. Pattern and predictors of mortality in necrotizing fasciitis patients in a single tertiary hospital. World J Emerg Surg. 2016;11: 40.

https://doi.org/10.1186/s13017-016-0097-y

27. Audureau E, Hua C, de Prost N, et al. Mortality of necrotizing fasciitis: relative influence of individual and hospital-level factors, a nationwide multilevel study, France, 2007-12. Br J Dermatol. 2017;177:1575-82.

https://doi.org/10.1111/bjd.15615