IDENTIFICATION OF POTENTIALLY MODIFIABLE FACTORS TO IMPROVE RECOGNITION AND OUTCOME OF NECROTIZING SOFT-TISSUE INFECTIONS

ABSTRACT Background: Necrotizing soft-tissue infections (NSTIs) present a surgical emergency of increasing incidence, which is often misdiagnosed and associated with substantial mortality and morbidity. A retrospective multicenter (11 hospitals) cohort study was initiated to identify the early predictors of misdiagnosis, mortality, and morbidity (skin defect size and amputation). Methods: Patients of all ages who presented with symptoms and were admitted for acute treatment of NSTIs between January 2013 and December 2017 were included. Generalized estimating equation analysis was used to identify early predictors (available before or during the first debridement surgery), with a significance level of P < 0.05. Results: The median age of the cohort (N = 216) was 59.5 (interquartile range = 23.6) years, of which 138 patients (63.9%) were male. Necrotizing soft-tissue infections most frequently originated in the legs (31.0%) and anogenital area (30.5%). More than half of the patients (n = 114, 54.3%) were initially misdiagnosed. Thirty-day mortality was 22.9%. Amputation of an extremity was performed in 26 patients (12.5%). Misdiagnosis was more likely in patients with a higher Charlson Comorbidity Index (β = 0.20, P = 0.001), and less likely when symptoms started in the anogenital area (β = −1.20, P = 0.003). Besides the established risk factors for mortality (septic shock and age), misdiagnosis was identified as an independent predictor of 30-day mortality (β = 1.03, P = 0.01). The strongest predictors of the final skin defect size were septic shock (β = 2.88, P < 0.001) and a skin-sparing approach to debridement (β = −1.79, P = 0.002). Conclusion: Recognition of the disease is essential for the survival of patients affected by NSTI, as is adequate treatment of septic shock. The application of a skin-sparing approach to surgical debridement may decrease morbidity.

Adequate treatment essentially consists of high-dose intravenous antibiotics, organ support, and sufficient surgical debridement (16), which can be performed skin-sparing (preserving all potentially viable skin) or "en bloc" (resecting all skin overlying the affected fascia) (16)(17)(18).Treatment should be initiated as soon as the diagnosis is made, as delays will negatively affect outcomes.Recognition is challenging because the rareness (0.4-1.4 per 100.000person years) (19,20) of NSTI and nonspecific initial symptoms, including pain, swelling, redness, and flu-like symptoms (21).
To improve the outcomes, potentially modifiable factors associated with mortality and morbidity should be identified.Previous studies have identified factors associated with sepsis and age that predict mortality (27).Other studies have identified American Association of Anaesthesiologists (ASA) classification (28) and cardiac comorbidity (29) to be associated with mortality.A systematic review concluded that delay until the first debridement surgery is related to mortality (30), which was confirmed by another study that found misdiagnosis to be associated with a delay to the first surgery (31).However, the relative strength of multiple patient, disease, presentation, and treatment characteristics that are available upon presentation has not previously been studied in a multivariable analysis to identify the strongest independent predictors of misdiagnosis, mortality, and morbidity (amputation and skin defect size).This was the aim of this study.

Study design
This study analyzed a subset of a Dutch retrospective multicenter cohort comprising 271 patients treated in 11 hospitals (4 university hospitals, 6 teaching hospitals, of which 3 have a burn center and 1 general hospital), and included a part (59 patients) of a previously described cohort (32).Patients admitted for acute treatment of NSTI between January 1, 2013, and December 31, 2017, were included.In this study, those who were referred after diagnosis and initial treatment in a nonstudy center were excluded.The medical ethics committee of Amsterdam University Medical Centre waived the requirement for consent.In all participating hospitals, approving was obtained at the relevant committee (institutional review board or local medical ethics committee).Reporting was performed according to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) criteria for cohort studies (33).

Patient identification and data collection
Patients were identified by a computer-generated search with the International Classification of Diseases, Tenth Edition, codes (34) (M72.6,A48.0, and N49.2) and free-text terms.In the burn centers, the Dutch Burn Centers Registry was used adjacently (35).All identified records were checked by one or two of the authors (JS, DH, LFW, AE).Initially, two authors performed selection, and results were compared.When sufficient agreement was reached, patient selection was continued individually.A similar approach was used for data extraction.Patients were included when the diagnosis was confirmed by unequivocal clinical or intraoperative findings or equivocal findings by a positive result on frozen section investigation.Patients readmitted for reconstructive surgery were excluded.Data on hospital, patient, disease, presentation, treatment characteristics, and outcomes were extracted from the medical records and collected in an electronic Case Report Form using Castor EDC.As a general rule, the earliest available data was used.Presentation characteristics (i.e., signs, symptoms, laboratory values) used were those that were obtained and/or described at the emergency department, or at first assessment in case of in-hospital developing NSTI.A complete overview of all the variables included can be found in Table 1, including the percentages of missing data.

Definitions
The American Society of Anesthesiologists (ASA) status (36), original Charlson Comorbidity Index (CCI) (37), and LRINEC score (23) were defined as in their original publications.Patients were defined as having septic shock when data in their medical records met the SEPSIS-3 criteria, or when a patient died upon presentation before adequate resuscitation could be performed (38).
Necrotizing soft-tissue infection was classified as in-hospital developed when one of the following three situations was true: (1) patients were admitted electively (e.g., planned surgery); (2) patients were admitted acutely for another diagnosis (e.g., cellulitis) that initially improved without surgery; or (3) surgery was performed initially for a nonnecrotizing soft tissue infection (e.g., drainage of an abscess) with no signs of NSTI during the surgery.
The duration of symptoms was defined as the number of days any symptoms (localized or systemic) led to the current presentation.An altered mental state was deemed present when the record mentioned that the patient was not as alert as usual.Late localized symptoms were defined as blue-livid discoloration, blisters, ecchymosis, or necrosis.A portal of entry was defined as any visible disruption of the skin or mucosa including an abscess.
Patients were classified as misdiagnosed when they were (1) admitted for treatment in the hospital due to symptoms that would later be proven to be due to NSTI, but this diagnosis was not considered, and an alternative diagnosis was made, or (2) in case of in-hospital NSTI (as defined previously) when symptoms developed, which would later be proven to be caused by NSTI, but this was not recognized upon assessment by a clinician, and an alternative diagnosis was made.The time to debridement was defined as the time between presentation and the first surgical debridement (hours).
Application of a skin-sparing approach was registered if this was explicitly mentioned in the operative note of the first debridement surgery or when the described procedure clearly stated preserving skin overlying the affected deep tissue layers.All other instances were registered as not being skin-sparing.
Amputation was registered without distinction between levels (distal or proximal).The final skin defect size was defined as the size of the resected skin area after the last surgical debridement and before reconstructive surgery, expressed as a percentage of the total body surface area (TBSA).Mortality (all-cause) was defined as death within 30 days of admission.

Statistical analysis
Statistical analysis was performed using IBM Statistical Package for the Social Sciences (SPSS) version 28.0 (IBM Corp., Armonk, NY).Categorical variables were described as frequencies with percentages, and continuous variables as medians with the interquartile range (IQR) represented by the lower and upper bounds (P25-P75).Missing data were defined as the frequency and percentage of the total cohort for which a certain characteristic was unknown.For statistical testing between groups, the Mann-Whitney U test (continuous data) or χ 2 test (categorical data) was used, in which missing data were excluded from the analysis.
Generalized estimating equation (GEE) analysis was used to identify factors associated with binary (mortality, amputation, misdiagnosis) or continuous (final skin defect size) outcomes, with the study center as the level.Generalized estimating equation analysis was performed on complete case data (listwise deletion) as well as on imputed data.A different imputed dataset was designed for each of the outcomes of interest, as graphically displayed in Supplementary Material Figures 1-4, http://links.lww.com/SHK/B893.Multiple imputation with predictive mean matching was performed and included variables with the highest correlation in the imputation models, to make the missing at random assumption plausible.Subsequent analyses were performed across the imputed datasets and Rubin's rules were used to summarize the model results.For the analysis of the final skin defect size, only those alive on day 30 were included (n = 162) to prevent the inclusion of patients in whom surgical debridement had not been completed.For the amputation analysis, only patients in whom an extremity was involved were included (n = 89), and one imputation model was constructed.For the other outcomes, two imputation models were constructed and merged.Backward elimination was used until only the significant predictors ( P < 0.05) remained.An overview of the variables included in the various GEE models is presented in Supplementary Material Int J Infect Dis Figures 1-4, http://links.lww.com/SHK/B893.
Outcomes of the GEE analysis were presented by their regression coefficients (β) (direction of the regression line and how much the dependent variable changed for each unit of change of the independent variable) and 95% confidence intervals, as well as odds ratios (ORs) for the logistic analysis.Outcomes of the analysis of the imputed datasets from the primary outcomes of this study.

RESULTS
The characteristics of the cohort (N = 216) are presented in Table 1.The median age was 59.5 (46.3-69.9)years.The majority of the patients were male (63.9%).One in four patients (25%) was considered obese (BMI >30 kg/m 2 ) and approximately half (50.5%) had severe systemic disease (ASA class 3 or 4).
A skin-sparing debridement approach was reported in 77 patients (35.6%).All patients received high-dose intravenous antibiotics according to local protocols, initially empirical, and tailored based on the culture results.Septic shock was present in 110 of the patients (52.1%) ; the majority (n = 102) met the SEPSIS-3 criteria, eight died before adequate resuscitation could be performed.Continuous venovenous hemofiltration (CVVH) was required in 23 patients (11.1%) due to acute kidney injury.Cultures revealed the presence of a group A Streptococcus (GAS) in 82 patients (39.4%).

Predictors for misdiagnosis
In Figure 1A, the outcome of the analysis of the predictors of misdiagnosis is graphically displayed, and Table 2 provides a detailed overview.In the primary (pooled) analysis, a higher CCI was positively associated with misdiagnosis (OR, 1.22; β = 0.20, P = 0.001).Compared with the development of NSTI in the lower extremity, development in the anogenital area was negatively associated with misdiagnosis (OR, 0.30, β = −1.20,P = 0.003).Both factors were identified as predictors in the complete case analysis.

Predictors of 30-day mortality
The factors positively associated with 30-day mortality in the primary analysis were septic shock (OR, 7.01, β = 1.95,  3 and Figure 1B.The presence of GAS was negatively associated with 30-day mortality (OR, 0.31, β = −1.17,P < 0.001).The complete case analysis only revealed age to be significantly associated with mortality.

Predictors of morbidity
As shown in Figure 1C and Table 4, septic shock was positively associated with an increased final skin defect size (β = +2.88,P < 0.001) in the primary analysis.Development of NSTI in the anogenital area (β = −1.77,P = 0.01) and a skin-sparing approach to debridement (β = −1.79,P = 0.002) were negatively associated with final skin defect size.Septic shock and a skin-sparing approach were also significantly associated with a complete case analysis.For amputation of an extremity, two negatively associated factors were identified in the primary analysis (Fig. 1D, Table 5): application of a skin-sparing approach (OR, 0.13, β = −2.05,P < 0.001) (also significant in the complete case analysis), and initial misdiagnosis (OR, 0.51, β = −0.68,P = 0.04).

DISCUSSION
This is the first multicenter study in which a broad variety of early predictors that may affect the outcomes of NSTI were included.The recent (twofold to threefold) increase in the incidence of NSTI caused by GAS in various European countries has increased the need for studies on the treatment and outcome of NSTI (39)(40)(41).The main finding of this study is the identification of misdiagnosis, which occurred in over half of the patients, as an independent predictor of 30-day mortality.This has not been previously described, although studies have reported increased mortality upon delay to the first surgery (30,31), and misdiagnosis is regarded as a main modifiable prognostic factor for mortality by clinicians (42).
Misdiagnosis is common worldwide (41%-96%) (22,43), thus, improving recognition could improve outcomes globally.Clinicians first need to realize that a high index of suspicion is essential, given the nonspecific initial symptoms of NSTI (pain, swelling, redness, and general sense of feeling ill) (21,22), and to understand which factors may predispose to misdiagnosis.Previously, a higher comorbidity status, absence of pain, and blue-livid skin discoloration were identified as factors associated with misdiagnosis (32).Only higher comorbidity status was confirmed to be associated with a higher risk of misdiagnosis in the current study.It is possible that patients with more (severe) comorbidities may present less typical, or a broader differential diagnosis may lead to NSTI being considered less likely.Although pain was not significantly associated with misdiagnosis in this study, the realization that pain may be absent in patients with NSTI remains relevant (5,44).The identification of anogenital NSTI (Fournier gangrene) as a factor associated with a decreased chance of misdiagnosis is not surprising.Fournier gangrene is a relatively well-known and well-described subtype of NSTI (45,46), due to which the diagnosis may be suspected sooner in patients presenting with anogenital skin changes and sepsis, compared with patients presenting with NSTI of other anatomic areas.
FIG. 1. Graphic overview of the significant ( P < 0.05) identified predictors (yellow text boxes) of each of the four outcomes (blue text boxes) explored.A (+) indicates a positive association between the predictor and outcome, meaning a positive (i.e., yes) or higher value of the predictor is associated with a positive or higher value of the outcome of interest.A (−) indicates the opposite; a positive or higher value of the predictor is associated with a negative (i.e., no) or lower value of the outcome of interest.More details on the strength of the associations can be found in Tables 2-5.Interestingly, the LRINEC score was found to be similar between patients initially misdiagnosed and those who were correctly diagnosed, making it unlikely to be of value in decreasing misdiagnosis.The explanation of why misdiagnosis is associated with an increased risk of mortality seems initially obvious, because an incorrect diagnosis will lead to a delay of the start of adequate treatment.However, although those misdiagnosed did face an extensive delay to first surgical debridement, the interval between presentation and first debridement surgery itself was not identified as an independent predictor of mortality in this study.One explanation could be that although delay to debridement will cause harm in all patients with NSTI, it will be more harmful in some patients compared with others.The speed of disease progression may differ considerably between patients and subtypes (47), and a delay of 2 h in patient 1 may be more harmful than 4 h in patient 2. Consequently, when the relative contribution of patients with slower disease progression is higher in those with a longer interval to first debridement, no association may be found.In contrast, misdiagnosis will lead to a delay in the start of adequate treatment in all patients regardless of subtype, and may therefore be a better predictor in a heterogeneous cohort.In addition, whereas time to first debridement focusses on surgery, misdiagnosis may be more clearly associated with a delayed start to adequate antibiotic treatment as well, thus having broader treatment implications.
Besides misdiagnosis, other factors found to be associated with 30-day mortality in the current study were septic shock and increased age, which is in line with the findings of previous work in which various characteristics associated with septic shock (low urine production, high lactic acid, hypotension) and age were highly predictive of 30-day mortality (27).A similar pattern was reported by Anaya et al. (48), who reported age above 50 years, and factors associated with sepsis (tachycardia, leukocytosis, hypercreatinemia), to be associated with increased mortality in patients with NSTI.The presence of GAS was associated with a decreased risk of mortality in the current study, which is in concordance with previous findings and may be due to patients infected by GAS having fewer comorbidities compared with patients infected with other pathogens (mostly polymicrobial) (5,49).
In addition to reducing mortality, it is highly relevant to decrease morbidity and improve the long-term quality of life.Morbidity caused by NSTI can be extensive and includes physical (e.g., scars, pain, fatigue), cognitive, psychological, social, and relational consequences (13)(14)(15).Two clinical outcomes measured and analyzed in this study (i.e., the size of the skin defect and amputation of an extremity) reflect physical morbidity.Septic shock was positively associated with the skin defect size.This is unsurprising because extensive tissue involvement indicates a more extensive spread of the disease and bacterial load.We believe that this emphasizes the importance of adequate treatment of sepsis in the ICU, which, together with surgical debridement and high-dose intravenous antibiotics, forms one of the three pillars of NSTI treatment (50).Although the analysis of amputation of an extremity showed two potentially modifiable factors that were associated (i.e., skin-sparing approach to debridement and initial misdiagnosis), we believe that no conclusions can be drawn based on these outcomes.It is unlikely that a skin-sparing approach lowers the chance of amputation, and more plausible that surgeons who use a skin-sparing approach are more willing to give a severely affected extremity the benefit of the doubt.It is also unlikely that an initial misdiagnosis would lead to a decreased amputation rate and that those initially misdiagnosed presented slightly different and had less severely affected limbs, resulting in poorer recognition and less amputation.
The strengths of this study are the multicenter design including different types of hospitals (i.e., university hospitals, teaching hospitals, burn centers, and general hospitals), as well as the broad search strategy performed, resulting in a representative cohort.Furthermore, a major strength is the inclusion of a broad variety of patient, disease, presentation, and treatment data, because multiple (potentially modifiable) factors could be tested in a single analysis.
The retrospective nature of this study is a limitation because the quality of the data is determined by the accuracy of the reported findings in patient records.The substantial amount of missing data for some of the collected data is another limitation of this study.This was dealt with by using multiple imputation, which can be viewed as the most reliable method to decrease bias and increase the accuracy of outcomes in case of missing data (51)(52)(53).Another potential limitation is the relatively large proportion of patients with NSTI of the anogenital and abdominal area, compared with some previous publications (43,(54)(55)(56)(57)(58).However, these studies are all single-center studies, and half of these studies (from Taiwan) had a high incidence of NSTI caused by Vibrio vulnificus and Aeromonas hydrophila (55,56,58).Because other studies found a similar distribution of affected areas (37.2%-39.6%anogenital, 7.0%-11.2%abdominal) (6,8), and the biggest prospective multicenter study on NSTI reported a similar 34% for anogenital and abdominal NSTI (5), we believe that our cohort is representative of the population of patients with NSTI in western countries.Furthermore, it is possible that susceptibility bias may have influenced the finding that a skin-sparing approach to debridement is associated with smaller final skin defects (59).This bias could be caused by surgeons selecting patients with certain prognostic factors (i.e., less extensive tissue involvement) to be treated with a skin-sparing approach.If this were the case, however, a similar effect of the skin-sparing approach on (lower) mortality would have been expected, because of surgeons selecting patients who are less severely ill to as skin-sparing approach.This association was not found, making it unlikely that susceptibility bias is a main explanatory factor of the association between surgical approach and skin defect size.
While this study provides valuable and relevant insights into the factors that may influence mortality and morbidity in patients with NSTI, we believe that these results need to be confirmed and knowledge expanded based on prospective studies.Studies should preferably include a standard set of uniformly defined variables, and clinical and patient-reported outcome measures.This may additionally reveal potentially modifiable factors that influence long-term morbidity, other than scar size and amputation.Ultimately, this may improve recognition and outcome.
In conclusion, NSTIs are commonly misdiagnosed, especially in patients with more (severe) comorbidities, and less often in cases of anogenital involvement.Misdiagnosis, increasing age, and septic shock are associated with increased 30-day mortality, and GAS is associated with lower mortality.Septic shock is also associated with larger final skin defects, and a skin-sparing approach to debridement is associated with smaller skin defects and less amputation.Thus, to improve outcomes, recognition, which requires a high index of suspicion, should be improved.This should be followed by adequate treatment of sepsis to reduce both mortality and morbidity and preferably a skin-sparing approach to debridement.

TABLE 1 .
Characteristics of the complete sample (N = 216) including respective rates of missing data *Missing data are not included in the calculation of percentages of categorical data.†Late symptoms: blue livid discoloration, blisters, ecchymosis, and necrosis.‡No missing data since all cases in which no skin-sparing approach was described were scored as not skin sparing.P < 0.001), increased age (OR, 1.05, β = +0.05,P = 0.002), and initial misdiagnosis (OR, 2.79, β = +1.03,P = 0.01), see Table

TABLE 2 .
Factors significantly associated with misdiagnosis for all included patients (N = 216), for both the analysis in the imputed dataset, as well as based on the complete case analysis NA, not applicable; NS = not significant.*compared with reference category leg.

TABLE 3 .
Factors significantly associated with 30-day mortality for all included patients (N = 216), for both the analysis in the imputed dataset, as well as based on the complete case analysis

TABLE 4 .
Factors significantly associated with final skin defect size for 30-day survivors (n = 162), for both the analysis in the imputed dataset, as well as based on the complete case analysis *Compared with reference category leg.NA, not applicable; NS, not significant.

TABLE 5 .
Factors significantly associated with risk of amputation of an extremity for patients in whom an extremity was involved (n = 89), for both the analysis in the imputed dataset, as well as based on the complete case analysis 590 SHOCK VOL.61, NO. 4