https://orcid.org/0000-0002-0661-9854
Figures
Abstract
Background
Snakebite is a neglected problem with a high mortality in India. There are no simple clinical prognostic tools which can predict mortality in viper envenomings. We aimed to develop and validate a mortality-risk prediction score for patients of viper envenoming from Southern India.
Methods
We used clinical predictors from a prospective cohort of 248 patients with syndromic diagnosis of viper envenoming and had a positive 20-minute whole blood clotting test (WBCT 20) from a tertiary-care hospital in Puducherry, India. We applied multivariable logistic regression with backward elimination approach. External validation of this score was done among 140 patients from the same centre and its performance was assessed with concordance statistic and calibration plots.
Findings
The final model termed VENOMS from the term “Viper ENvenOming Mortality Score included 7 admission clinical parameters (recorded in the first 48 hours after bite): presence of overt bleeding manifestations, presence of capillary leak syndrome, haemoglobin <10 g/dL, bite to antivenom administration time > 6.5 h, systolic blood pressure < 100 mm Hg, urine output <20 mL/h in 24 h and female gender. The lowest possible VENOMS score of 0 predicted an in-hospital mortality risk of 0.06% while highest score of 12 predicted a mortality of 99.1%. The model had a concordance statistic of 0·86 (95% CI 0·79–0·94) in the validation cohort. Calibration plots indicated good agreement of predicted and observed outcomes.
Author summary
More than 58,000 people die of snakebites each year in India. Russell’s viper, saw scaled viper and pit vipers are widely distributed and medically important venomous snakes in India responsible for significant deaths and disabilities. Clinicians need easy-to-use bedside tools to make decisions on which patients are at a higher risk of dying after viper bites. In this study, conducted in Southern India, where Russell’s viper is the commonest viper causing bites, we have evolved and validated a simple risk prediction score. This uses seven clinical and laboratory features to estimate a patient’s risk of dying in the hospital due to the bite. The study showed that the score has good predictive ability when tested in a similar population of patients. We expect that the score is the first step in developing a tool that is likely to help health workers and doctors assess a patient’s risk in primary-care peripheral or rural settings to help decide on early referral of high-risk patients who are likely to worsen.
Citation: Gopalakrishnan M, Saurabh S, Sagar P, Bammigatti C, Dutta TK (2022) A simple mortality risk prediction score for viper envenoming in India (VENOMS): A model development and validation study. PLoS Negl Trop Dis 16(2): e0010183. https://doi.org/10.1371/journal.pntd.0010183
Editor: Nicholas R. Casewell, Liverpool School of Tropical Medicine, UNITED KINGDOM
Received: June 3, 2021; Accepted: January 20, 2022; Published: February 22, 2022
Copyright: © 2022 Gopalakrishnan et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: The authors received no funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Snakebite envenoming is a serious but neglected problem in the tropics [1,2]. South Asia, particularly, India has the largest burden of snakebite deaths and disabilities in the world [3,4]. Recent estimates suggest that annual mortality from snakebite envenomings in India is approximately 58,000; which is more than half the estimated global snakebite mortality. Thrice as many endure lifelong disabilities due to long-term consequences [4,5]. Affected are usually young adults belonging to lower socio-economic background who experience subsequent social stigma and discrimination [4,6,7]. In South Asia, the snake species under the epithet of “Big 4” i.e. Daboia russelii, Echis carinatus, Bungarus caeruleus and Naja naja garner widespread attention while the other regionally important snake species are also emerging as medically important [8,9]. The currently available polyvalent antivenom in India neutralizes venom of only these four species [10].
In clinical settings, snakebite envenoming syndromes are broadly categorized as neurotoxic and haemo-vasculotoxic. Neurotoxic symptoms are usually due to elapid bites i.e., cobra and krait, and vasculotoxic envenomings due to vipers. An important bedside test in establishing the diagnosis of viper envenoming is the whole blood clotting test (WBCT20) [11]. Two ml of freshly sampled venous blood in a dry, glass vessel or tube and left undisturbed for 20 minutes at ambient temperature. The vessel is tipped once, if the blood is still liquid (unclotted) and runs out, the patient is inferred to have hypofibrinogenaemia (“incoagulable blood”) as a result of venom-induced consumption coagulopathy (VICC) [11].
Among the viper species, Russell’s viper is widely distributed throughout Indian subcontinent including Sri Lanka and Myanmar [12,13]. A nationwide study and evidence synthesis estimated that 43% of reported bites in India are likely to be due to Russell’s viper envenoming [4]. Russell’s viper is reported as the species responsible for up to 80% mortality in several hospital-based series across India [14,15]. Variation in venom composition between Russell’s viper species from various parts of India leading to marked difference in neutralizing capability of the polyvalent antivenom has been recently demonstrated [12].
Russell’s viper envenoming is clinically complex and challenging as it results in a rapidly progressive multisystem dysfunction culminating in mortality. The envenoming haemo-vasculotoxic syndrome affects platelets, coagulation factors (like factors V and X), endothelium of the vessel wall resulting in VICC, thrombotic microangiopathy and capillary leak syndrome (CLS) [16,17]. VICC presents with bleeding manifestations which can range from mild bleeding like gum bleeds and bite-site bleeding to life threatening bleeds such as intracranial haemorrhage and gastrointestinal bleeds [18]. CLS has been reported from Russell’s viper bites in Southern India, Sri Lanka and Myanmar and is associated with a poor outcome [10,19]. CLS presents with manifestations of parotid swelling, conjunctival-chemosis, periorbital edema, hypotension, albuminuria and hemo-concentration. Other organ-systems including kidneys, heart, presynaptic neuromuscular junction and hypothalamo-pituitary axis are also affected in Russell’s viper envenoming resulting in acute kidney injury (AKI), early neuromuscular paralysis, acute adrenal insufficiency and long-term consequences like chronic kidney disease and Sheehan like syndrome [20–24].
The other important viper species with widespread distribution is Echis carinatus which has two subspecies: Echis carinatus accounts for for envenomings in the Indian peninsula while Echis carinatus sochureki is thought to be responsible for bites in Northern India and Pakistan [25,26]. Echis envenoming presents with local swelling, coagulopathy and bleeding manifestations. Apart from this several pit vipers such as hump-nosed pit viper (Hypnale hypnale), Himalayan and bamboo pit vipers in north-eastern India and Malabar pit viper in the western coast are also clinically significant [10]. Hump nosed pit viper can cause local necrosis, coagulopathy, bleeding and acute kidney injury and maybe misidentified as saw-scaled viper [27,28]. Syndromic diagnosis is widely applied especially in primary care settings despite its limitations in the absence of reliable species identification methods in routine clinical practice [29,30].
Managing viper bites is complicated involving multiple decisions like need for renal replacement therapy, ventilatory support for pulmonary edema, ionotropic support for distributive shock in capillary leak syndrome and transfusion support based on which organ systems are involved and when [31]. This is supported by several studies which report higher mortality and morbidity in viper envenoming [4,32]. Thus, viper envenomings have a complex pathogenesis with distinct prognostic factors involved implying that they merit the need for a distinct clinical decision support tool from elapid envenoming.
Recently, World health organization (WHO) has evolved a strategy to halve the snakebite mortality by 2030 as compared to 2015. One of the strategies in this call to action includes development of clinical decision support tools for improving outcomes [33]. Though several clinical parameters have been explored as mortality risk predictors in hospital-based studies, no simple score exists to quantify the prognostic factors affecting outcomes [34,35].
We aimed to develop and externally validate a simple, point-of-care mortality risk prediction score for patients presenting with syndromically diagnosed hemotoxic viper envenoming patients which could be potentially applied across healthcare settings.
Methods
Ethics statement
Both studies were approved by institutional ethics committees (JIPMER Institue Ethics Committee, JIP/IEC/SC/3/2012/13 and JIP/IEC/2014/1/24). Written informed consent was obtained at the time of data collection from the participant or parent/guardian along with participant’s assent if the participant’s age <18 years. However, repeat consent was not obtained, as this was a retrospective study using de-identified patient data from previous studies.
Study setting, populations, and design cohorts: The study site for both derivation and validation cohorts was a tertiary care referral hospital situated in Puducherry, India, located in the eastern coast of India. Our hospital has a catchment area of approximately 17,000 km2 wherein 8 medically important snakes including the “big 4” have been routinely reported [36]. (Fig 1). Russell’s viper is common and saw-scaled viper is routinely reported while pit vipers have not been reported in the area.
The union territory of Puducherry (town), showing location of the study site with highlighted adjacent districts of the state of Tamil Nadu from where patients were enrolled. (Map not to scale. Maps created using https://www.datawrapper.de/).
Derivation cohort
We developed the model using data from a prospective cohort study of consecutive patients presenting to the emergency department of a tertiary care referral centre in Puducherry, India between September 2011 to August 2013.The clinical characteristics and outcomes of this prospective derivation cohort (n = 248) have been published previously [37]. Those patients ≥ 12 years of age, presenting with a history of snakebite or unknown bite with positive whole blood clotting test (WBCT20) and diagnosis of viper envenoming made by syndromic approach or identification of dead snake/photograph of the snake if brought by the patient were included. Syndromic diagnosis of viper envenoming was made based on syndromes 1, 2 and 5 in World Health Organization (WHO) guidelines. Syndrome 1 (All viperidae): Local envenoming (swelling) with bleeding/clotting disturbances. Syndrome 2: (Russell’s viper in South India/Myanmar/Sri Lanka): Local envenoming and bleeding/clotting disturbances with shock, acute kidney injury, conjunctival chemosis, acute pituitary insufficiency, ptosis, external ophthalmoplegia, facial paralysis or dark brown urine. Syndrome 5 (Russell’s viper in Sri Lanka or South India): Bitten on land and paralysis with dark brown urine/acute kidney injury with bleeding/clotting disturbances. Those with isolated neurotoxicity and local manifestations alone with normal WBCT20 were excluded (Syndromes 3 and 4) [31]. All patients in this cohort presented within 48 hours of bite while 67% presented within 6 hours of bite.
Validation cohort
We validated the model in an external cohort of 140 patients who presented to the same centre from September 2013 to July 2015.This cohort was comprised of patients from a randomized clinical trial investigating two different doses of polyvalent antivenom [38]. This cohort included patients who had abnormal WBCT20 and syndromic diagnosis of viper envenoming. However, this cohort excluded those who had received greater than 200 mL (20 vials) antivenom prior to presentation (trial registered at CTRI/2015/05/005826). All patients in this cohort also presented within 48 hours of bite.
Predictor variable selection
We searched for predictors of mortality in haemotoxic viper bite envenoming that were reported in previous studies or reviews (Table A in S1 Appendix -). We selected parameters that could easily be ascertained in different clinical settings with minimal interobserver variability and were part of the routine assessment in snakebite envenoming especially in primary care settings. Coagulation tests such as prothrombin time (PT), activated partial thromboplastin time (aPTT), serum fibrinogen, D-dimer were deliberately omitted considering the poor availability of these tests as point-of-care in primary care rural settings in India. For the purpose of this study clinical parameters assessed at 24 hours of admission, were defined as follows: a) signs of capillary leak syndrome (CLS) was defined as the presence of clinical evidence of at least one of the following: conjunctival chemosis, parotid swelling or periorbital puffiness with clinical evidence of pleural effusion or ascites b) overt bleeding: presence of bleeding from oral cavity, persistent bleeding from bite site hematuria, epistaxis, bleeding from intravenous puncture sites, hematemesis or melena, fresh bleeding per rectum, abnormal uterine bleeding or intracranial hemorrhage. c) renal dysfunction: Arbitrarily defined as serum creatinine > 3.0 mg/dl.) d) severe local envenoming: swelling involving more than one half of the bitten limb and bites involving the face/trunk. Urine output was measured over first 24 hours of admission and later converted to ml/hour.
Receiver operator characteristics (ROC) curves were constructed for each of the continuous variables from the derivation cohort to determine appropriate cut-offs to categorize them into clinically significant categories (Table B in S1 Appendix). Categorization of continuous variables was done in order to simplify the final score. For identifying additional predictors, we performed univariable (unadjusted) logistic regression analysis for each of identified risk factors and few others as dependent variables with mortality as outcome and we included significant (p < 0·05) predictors for model development (Table C in S1 Appendix). Sample size estimation was done using a thumb rule of 10 events per predictor [39]. As there were 57 events in the derivation cohort, the ideal number for predictors in the model was taken to be 6 to 7. Multiple imputation analysis was planned for addressing missing data if missing data for any predictor>5%.
Model development
The predictors finally selected for the multivariable model are enumerated in Table C in S1 Appendix. All candidate variables from the derivation cohort were entered into the multivariable logistic regression analysis. We used a backward stepwise elimination approach with the least statistically significant variable removed at each step. A total of five elimination steps simplified the model based on minimum Akaike Information Criteria (AIC) value.
Conversion to score
In the final model, we assigned the scores proportional to their β regression coefficients of the multivariable regression equation, using standard approach [40]. The variable with minimum β value was assigned a score of 1 and the remaining variables were assigned proportional scores with rounding off to the nearest integer to generate an easily calculable score [39,40]. An arbitrary cut-off score was chosen based on the score-mortality estimate graph.
Model performance, predictive accuracy, and external validation
Discrimination (i.e., the degree to which a model differentiates between those who died and survived) was calculated with concordance (c-index or statistic), equivalent to the area under the ROC curve. A value of 0.5 indicates no predictive ability, 0.8 is considered good, while 1 is perfect discrimination. Hosmer and Lemeshow goodness of fit statistic and Nagelkerke r2 were calculated for assessing overall model performance. To assess the calibration of the model, (i.e., agreement between predicted and observed risk of mortality), calibration plots were used. Perfect calibration is implied by a 45° diagonal line (calibration slope = 1 and a calibration intercept = 0). Deviations above or below the line reflects underprediction and overprediction by the model. We assessed the predictive accuracy of the score in the validation cohort with discrimination and calibration as mentioned above. We did all analysis with SPSS statistical software v23. Calibration plots were constructed Stata/IC v16 (trial version). The present study is reported in compliance with standard TRIPOD guidelines for prediction models (S1 TRIPOD Checklist).
Results
For the selection of candidate variables, 15 studies were reviewed to generate a list of 25 potential parameters. Related parameters were combined for clarity (e.g., shock and hypotension, anaemia, and haemoglobin < 10 g/dL). Ten parameters were considered infeasible for primary care settings and were excluded, among which, 3 were not deemed suitable for measurement on day 1 of bite. Two parameters reported in only a single study done on children were also not included (Table A in S1 Appendix). The derivation cohort included 248 while the validation cohort comprised 140 participants. Baseline characteristics for both cohorts are summarized in Table 1.
In the derivation cohort, 74.1% (n = 184) and validation cohort, 79.2% (n = 119) were classified as Russell’s viper envenoming by either snake identification or syndromic diagnosis (syndromes 2 & 5). Also, 19% in derivation cohort and 15% in validation cohort were classified as viper envenoming with unspecified species—syndrome 1 i.e., local swelling with prolonged WBCT20. A section of these patients is also expected to be Russell’s viper envenoming.
Univariable analysis in the derivation cohort (Table B in S1 Appendix,) found a significant association of in-hospital mortality with several predictors that were consistently reported previously: systolic blood pressure <100 mm Hg, presence of signs of capillary leak syndrome (CLS), any overt bleeding manifestations at admission, severity of local swelling, bite-to-antivenom time> 6.5h, haemoglobin <10 g/dL, presence of acute kidney injury (defined as creatinine >3 mg/dL), urine output < 20 mL/hour in the first 24 hours (measured over 24 hours), urine albumin positive by dipstick and thrombocytopenia (platelet < 260 x 109/L) (Table 2). These variables were entered into a multivariable model. Age and gender were also included in the model, despite being non-significant in the univariable analysis, because they were clinically relevant predictors.
Seven predictors remained in the multivariable model at step 5: overt bleeding, haemoglobin at admission <10 g/dL, bite to antivenom time> 6.5 hours, systolic blood pressure at admission < 100 mm Hg, presence of signs of capillary leak syndrome, urine output < 20 mL/hour in the first 24 hours and female gender (Tables 3 and 4). The predictors which were not significant at step 5 were also retained in the model considering optimal AIC and need to retain some clinically important predictors like bite-to-antivenom time which clinicians find valuable. Although AIC was minimum in step 6, we limited to five elimination steps in order to retain bite-to-antivenom time a clinically significant predictor variable as mentioned above based on clinician inputs and prior reports[34]. (Table 2 and Tables C and D in S1 Appendix). The regression equation and intercept (baseline mortality risk) are shown in Table 4. We assigned point values to these items and developed an integer-based estimation system (Tables 2 and 3).
Missing data
Missing data was < 5% for the predictor variables as data collection was prospective in the derivation cohort. Of the relevant predictors, data were 99·1% complete for 2 predictors (haemoglobin, platelet count) and 97.8% for serum creatinine. Data were complete for 100% of outcome parameters in the derivation cohort. Data was 100% complete for predictors and outcomes in the validation cohort as it was a randomized trial. As missing data was <5% we did not perform multiple imputation analysis.
Internal validation, discrimination, and calibration
Mortality risk plotted against each point of the score showed a sigmoid curve with steep increase in mortality when score was greater than 6 (Fig 2A). Hence, we decided to take a score of 6 as a cut-off for poor prognosis. Model discrimination using a ROC showed Area Under Curve (AUC/c-index) of 0.948 (95% CI 0.92–0.98) suggesting excellent discrimination. A cut-off of 6 as discussed above had a sensitivity of 90% and specificity of 83% for predicting mortality (Fig 3A). Hosmer-Lemeshow goodness of fit showed a chi-squared statistic of 1.52 (p = 0.99, df = 8) suggesting a good model fit. Nagelkerke r2 at step 5 was 0.69 again suggesting that the model explained 70% of the variability in the outcome parameter and a good overall performance (Table D in S1 Appendix). Internal calibration showed a slope of 1, intercept of 0 and an AUC of 0.95 suggesting excellent calibration in the derivation dataset (Fig A in S1 Appendix).
A: Mortality risk plotted against each point of the score for the derivation cohort (n = 248) showing a sigmoid curve with steep increase in mortality at score was greater than 6. B: Mortality prediction estimates for validation cohort (n = 140).
A: Model discrimination in derivation cohort using a receiver operator characteristic curve (ROC) showing area Under Curve (AUC/c-index) of 0.948 (95% CI 0.920–0.976). A cut-off of 6 had a sensitivity of 90% and specificity of 83% for predicting mortality. B: Model performance in validation cohort using a ROC showing AUC/c-index of 0·90 (95% CI 0·85–0·97).
External validation
The score was a significant predictor of mortality in the validation cohort (Odds ratio [OR] 1·8 per unit increase in score, 95% CI; p < 0·0001). Model performance in the validation cohort showed a c-statistic of 0·90 (95% CI 0·85–0·97) (Fig 2B). The model predicted a mean probability of mortality as 11% (95% CI 8–15%) in the validation cohort. Thus the 95% CI included the actually observed mortality of 14.3% indicating that calibration at large was satisfactory. Calibration plots of predicted and observed mortality showed a slope of 0.7, intercept of 0.4 and a c-index (AUC) of 0.92 suggesting overall overfitting of the model within the validation cohort with overprediction at low-risk patients and underprediction of mortality in high-risk patients (Fig 4). Prediction estimates in validation cohort are shown in Fig 2B. In the validation cohort, the lowest score of 0 predicted a mortality risk of 0.06% while a score of 12
predicted a mortality of 99.1%. Sensitivity, specificity positive and negative predictive values (PPV and NPV) at each point in the score was calculated for the validation cohort and is presented in Table 5. At the selected cut-off of 6 the sensitivity was 75%, specificity 88.3%, PPV 52% and NPV 96% in the validation cohort.
Fig 2B: Mortality prediction estimates for validation cohort (n = 140).
Fig 3B: Model performance in validation cohort using a ROC showing AUC/c-index of 0·90 (95% CI 0·85–0·97).
Calibration plots showing a slope of 0.7, intercept (CITL) of 0.4 and a c-index (AUC) of 0.92. E:O: ratio of expected to observed mortality. Graph created using pmcalplot in STATA, Stata/IC 16 for Windows.
Discussion
Snakebite envenoming usually affects those living in rural areas and in poverty [1,2,6]. Ending this neglect requires a refocus of research efforts into various aspects of snakebite envenoming including prognostic models to help classify patients according to severity and help plan appropriate management.
In this study, we have developed a practical prognostic instrument to predict the risk of in-hospital mortality after viper envenoming. The VENOMS score calculated on the day of admission was successfully externally validated and showed good discrimination and reasonable calibration in the same settings. The model incorporates seven items: overt bleeding manifestations, presence of signs of capillary leak syndrome, systolic blood pressure <100 mm Hg, urine output < 20 mL/h over first 24 hours (assessed over 24 hours), haemoglobin <10 g/dL, female gender, and bite to ASV time >6.5 hours. We prudently selected a list of candidate predictors and categorized them in the derivation cohort. Such a process involves making compromises, such as the exclusion of parameters that are not routinely assessed in a primary care clinical setting or that are not supported by sufficient validation data. The derivation cohort was adequately powered to show a good discrimination of the model. This is indicated by the 95% CIs of concordance statistics, which exceeded 0.8 in this cohort. Development and validation of the score followed established TRIPOD recommendations [41].
Prognostic scores support and improve the clinical decision making process and impact care by empowering clinicians to make evidence based decisions thereby improving patient outcomes[39]. Classical examples include Wells score for predicting pulmonary embolism and CURB 65 or pneumonia severity index for community acquired pneumonia. Both these scores have gained widespread applicability and have resulted in impacting diagnosis and management of these conditions including reduction in mortality of admitted patients in emergency departments [42,43]. Limited clinical prediction scores are available for neglected tropical diseases [44] A commonly reported score for snakebites is the Snakebite Severity Score (SSS) which ranges from 0 to 23 and assesses respiratory, cardiovascular, hematologic, gastrointestinal, central nervous system and local wound to assign scores for each [45]. The SSS was originally evolved for evaluating dry bites and deciding if patient requires antivenom or not. SSS has been shown to limit antivenom and other resource utilization [46,47]. It has been used as a prognostic score for haemotoxic bites in Indian settings, but a formal validation is unavailable [48]. The SSS has several limitations: it combines both neurotoxic and hemotoxic manifestations, includes several laboratory results including PT, aPTT, serum fibrinogen which are usually not available at primary care settings and common elapid neurological signs like ptosis do not figure in the score [49]. Apart from the SSS, studies from Korea have used the International Society of Thrombosis and Haemostasis scoring system for disseminated intravascular coagulation to classify viper bite patients with VICC though prognostic implications were unclear [50,51]. Another prognostic score is the Zululand Severity Score developed in South Africa for determining whether the patient requires antivenom or surgical intervention [52]. A species-specific severity grading for Indian snakes was evolved by Kumar V et al and was reported in subsequent hospital based studies [53,54]. However, the score is complex, the basis for severity grading are unclear and its prognostic implications were not validated. Patient-Specific Functional Scale (PSFS), is a patient-reported outcome that is validated for assessing limb recovery from snakebite envenoming [55]. In summary, there exists a need for a simple bedside prognostic instrument which can help triage and appropriately manage viper envenoming patients.
The VENOMS score has several potential practical applications despite being currently validated in a single centre: it can be applied readily at the bedside by clinicians without any device to stratify viper envenoming patients. We expect that the score can help tailor care according to risk-class by triaging low and high mortality risk (score >6) patients who may require early intensive care. We hypothesize that the score might aid decision making for early transfers while reducing unnecessary referrals in primary care settings. We also suspect that the score has a potential to reduce antivenom overuse in the form of additional doses in patients with low VENOMS score (e.g., a cut-off < 4 have mortality of 1.5%) similar to the SSS [46]. However, further clinical studies are warranted to confirm these suggestions. Cost-effectiveness and acceptability of VENOMS score also need further research. Likewise, the study opens several interesting questions which need further exploration in clinical context such as what are appropriate measures to reduce mortality, in high-risk individuals (Score >6) and what is performance of the score as a guide to supportive care?
Our study has several important limitations. A syndromic approach to identifying the offending snake may have resulted in errors. The scoring system has only been validated in the same centre as the derivation cohort, where the common species is Russell’s viper (at least 74% patients in derivation cohort and 79% in validation cohort fitted into confirmed or syndromic diagnosis of Russell’s viper). The score requires independent external validation in other settings before widespread applicability. The performance of this score in settings where saw-scaled viper envenoming forms bulk of cases will need appropriate modification of the score. The scope of the score is limited to in-hospital mortality.
Clinical manifestations vary greatly across India and South Asia, and our sample is from a single site. Geographical intraspecific variations in Russell’s viper envenoming has been known to cause varied clinical manifestations [12]. For example, capillary leak syndrome due to Daboia russelii envenoming has been frequently reported from Southern India, Sri Lanka, and Myanmar while there are only few reports of this phenomenon in from other areas in the subcontinent [19]. Likewise, pre-synaptic neurotoxic features in Russell’s viper envenoming appear to have limited geographical distribution [23]. Therefore, apart from the spectrum effect in clinical prediction scores, the score requires further widespread geographical as well as domain validation specifically in primary care settings.
All predictors were converted to categorical variables for ease of use, this might have led to some loss of information. There were some differences in baseline characteristics of both the cohorts even though they were from the same centre. This difference could be attributed to differences in study design (prospective cohort vs randomized clinical trial) and inclusion and exclusion criteria. Specifically, the validation cohort excluded patients who had received > 20 vials antivenom prior to admissions. It is possible that some severely envenomed patients (who are likely to receive higher doses of antivenom upfront at primary care settings) were missed in the derivation cohort. Also, even though both cohorts received antivenom from the same manufacturer (Table 1), multiple batch numbers were used according to institutional supply which might have resulted in varying action due to batch to batch variation [56,57]. It is pertinent to note that the median antivenom dose used by the derivation cohort is 30 vials which is the recommended upper limit for Russell’s viper envenoming suggesting that many patients received more antivenom than recommended but did not respond as expected. Also, the results are only applicable to adults >12 years of age as we did not include children who may have different clinical predictors as suggested by previous studies.
Selection bias needs to be considered because both cohorts pre-selected people with severe envenoming and the population was a tertiary care referral centre [39]. Both cohorts used clinical syndromic approach to snake identification based on the current WHO guidelines while serum-based assays could have ascertained species-based diagnosis of viper envenoming. However, this approach mimics a real-life situation, including rural primary care scenarios, possibly making the model applicable in these practice settings. There was deviation from the perfect slope in validation calibration plot (Fig 4). These deviations were limited in scope and within the estimated 95% CI. Also, smoothing techniques used to estimate the observed probabilities of the outcome in relation to the predicted probabilities, i.e. the loess algorithm may have affected the graphical impression, considering that the derivation cohort is a smaller dataset [58].
In conclusion despite limitations, the VENOMS score appears to be an easy-to-use point of care clinical prediction score for mortality prediction for Russell’s viper envenoming in Southern India with potential widespread applications in various settings.
Supporting information
S1 TRIPOD Checklist. TRIPOD checklist for prediction model development.
https://doi.org/10.1371/journal.pntd.0010183.s001
(DOCX)
S1 Appendix Text. Search Strategy, Potential variables considered and references.
Table A: Publications screened for variable selection for model development Table B: (Supplementary Appendix 1): Area under the curve (AUC) for Receiver-operating curves (ROC) constructed for continuous predictor variables with mortality (or survival) as the state variable with confidence interval (CI), cut off chosen and sensitivity and specificity at chosen cut-off. Table C: Odds ratio with 95% CI for univariable Binary Logistic Regression (unadjusted) and subsequent multivariable logistic regression with backward elimination strategy (adjusted) to predict mortality as outcome. Table D: Multivariable logistic regression model with backward elimination at step 5 and step 7 Table E: Model summary showing -2 log likelihood, Cox and Snell’s R square, Nagelkerke R Square and Akaike Information criteria (AIC) shown for each step of backward elimination. Fig A: Perfect Internal calibration in derivation cohort (slope of 1, intercept of 0 and an AUC of 0.95). Graph created using pmcalplot in STATA, Stata/IC 16 for Windows.
https://doi.org/10.1371/journal.pntd.0010183.s002
(DOCX)
S1 Data. Deidentified patient data for derivation cohort.
https://doi.org/10.1371/journal.pntd.0010183.s003
(PDF)
S2 Data. Deidentified patient data for validation cohort.
https://doi.org/10.1371/journal.pntd.0010183.s004
(PDF)
Acknowledgments
We gratefully acknowledge Dr L. Jeyaseelan, Professor and Head, Department of Biostatistics, Christian Medical College, Vellore for his guidance in multivariable logistic regression and score assignment.
References
- 1. The Lancet null. Snakebite-emerging from the shadows of neglect. Lancet Lond Engl. 2019 Jun 1;393(10187):2175. pmid:31162065
- 2. Babo Martins S, Bolon I, Chappuis F, Ray N, Alcoba G, Ochoa C, et al. Snakebite and its impact in rural communities: The need for a One Health approach. PLoS Negl Trop Dis. 2019 Sep;13(9):e0007608. pmid:31557159
- 3. Kasturiratne A, Wickremasinghe AR, de Silva N, Gunawardena NK, Pathmeswaran A, Premaratna R, et al. The global burden of snakebite: a literature analysis and modelling based on regional estimates of envenoming and deaths. PLoS Med. 2008 Nov 4;5(11):e218. pmid:18986210
- 4. Suraweera W, Warrell D, Whitaker R, Menon G, Rodrigues R, Fu SH, et al. Trends in snakebite deaths in India from 2000 to 2019 in a nationally representative mortality study. eLife. 2020 Jul 7;9.
- 5. Bhaumik S, Gopalakrishnan M, Meena P. Mitigating the chronic burden of snakebite: turning the tide for survivors. Lancet Lond Engl. 2021 Oct 16;398(10309):1389–90. pmid:34537105
- 6. Harrison RA, Hargreaves A, Wagstaff SC, Faragher B, Lalloo DG. Snake envenoming: a disease of poverty. PLoS Negl Trop Dis. 2009 Dec 22;3(12):e569. pmid:20027216
- 7. Weiss MG. Stigma and the social burden of neglected tropical diseases. PLoS Negl Trop Dis. 2008 May 14;2(5):e237. pmid:18478049
- 8. Simpson ID, Norris RL. Snakes of medical importance in India: is the concept of the “Big 4” still relevant and useful? Wilderness Environ Med. 2007;18(1):2–9. pmid:17447706
- 9. Senji Laxme RR, Khochare S, de Souza HF, Ahuja B, Suranse V, Martin G, et al. Beyond the “big four”: Venom profiling of the medically important yet neglected Indian snakes reveals disturbing antivenom deficiencies. PLoS Negl Trop Dis. 2019 Dec;13(12):e0007899. pmid:31805055
- 10. Warrell DA, Gutiérrez JM, Calvete JJ, Williams D. New approaches & technologies of venomics to meet the challenge of human envenoming by snakebites in India. Indian J Med Res. 2013;138:38–59. pmid:24056555
- 11. Lamb T, Abouyannis M, de Oliveira SS, Shenoy K R, Geevar T, Zachariah A, et al. The 20-minute whole blood clotting test (20WBCT) for snakebite coagulopathy-A systematic review and meta-analysis of diagnostic test accuracy. PLoS Negl Trop Dis. 2021 Aug;15(8):e0009657. pmid:34375338
- 12. Senji Laxme RR, Khochare S, Attarde S, Suranse V, Iyer A, Casewell NR, et al. Biogeographic venom variation in Russell’s viper (Daboia russelii) and the preclinical inefficacy of antivenom therapy in snakebite hotspots. PLoS Negl Trop Dis. 2021 Mar;15(3):e0009247. pmid:33764996
- 13. Kalita B, Mackessy SP, Mukherjee AK. Proteomic analysis reveals geographic variation in venom composition of Russell’s Viper in the Indian subcontinent: implications for clinical manifestations post-envenomation and antivenom treatment. Expert Rev Proteomics. 2018 Oct;15(10):837–49. pmid:30247947
- 14. Kumar KS, Narayanan S, Udayabhaskaran V, Thulaseedharan NK. Clinical and epidemiologic profile and predictors of outcome of poisonous snake bites—an analysis of 1,500 cases from a tertiary care center in Malabar, North Kerala, India. Int J Gen Med. 2018;11:209–16. pmid:29892202
- 15. Ghosh R, Mana K, Gantait K, Sarkhel S. A retrospective study of clinico-epidemiological profile of snakebite related deaths at a Tertiary care hospital in Midnapore, West Bengal, India. Toxicol Rep. 2018;5:1–5. pmid:29234603
- 16. Maduwage K, Isbister GK. Current treatment for venom-induced consumption coagulopathy resulting from snakebite. PLoS Negl Trop Dis. 2014 Oct;8(10):e3220. pmid:25340841
- 17. Udayabhaskaran V, Arun Thomas ET, Shaji B. Capillary Leak Syndrome Following Snakebite Envenomation. Indian J Crit Care Med Peer-Rev Off Publ Indian Soc Crit Care Med. 2017 Oct;21(10):698–702. pmid:29142382
- 18. Berling I, Isbister GK. Hematologic effects and complications of snake envenoming. Transfus Med Rev. 2015 Apr;29(2):82–9. pmid:25556574
- 19. Kendre PP, Jose MP, Varghese AM, Menon JC, Joseph JK. Capillary leak syndrome in Daboia russelii bite-a complication associated with poor outcome. Trans R Soc Trop Med Hyg. 2018 01;112(2):88–93. pmid:29584906
- 20. Warrell DA. Snake bite. Lancet Lond Engl. 2010 Jan 2;375(9708):77–88.
- 21. Herath HMNJ, Wazil AWM, Abeysekara DTDJ, Jeewani NDC, Weerakoon KG a. D, Ratnatunga NVI, et al. Chronic kidney disease in snake envenomed patients with acute kidney injury in Sri Lanka: a descriptive study. Postgrad Med J. 2012 Mar;88(1037):138–42. pmid:22282736
- 22. Antonypillai CN, Wass J a. H, Warrell DA, Rajaratnam HN. Hypopituitarism following envenoming by Russell’s vipers (Daboia siamensis and D. russelii) resembling Sheehan’s syndrome: first case report from Sri Lanka, a review of the literature and recommendations for endocrine management. QJM Mon J Assoc Physicians. 2011 Feb;104(2):97–108. pmid:21115460
- 23. Silva A, Kuruppu S, Othman I, Goode RJA, Hodgson WC, Isbister GK. Neurotoxicity in Sri Lankan Russell’s Viper (Daboia russelii) Envenoming is Primarily due to U1-viperitoxin-Dr1a, a Pre-Synaptic Neurotoxin. Neurotox Res. 2017;31(1):11–9. pmid:27401825
- 24. Bhattacharya S, Krishnamurthy A, Gopalakrishnan M, Kalra S, Kantroo V, Aggarwal S, et al. Endocrine and Metabolic Manifestations of Snakebite Envenoming. Am J Trop Med Hyg. 2020 Oct;103(4):1388–96. pmid:32602439
- 25. Alirol E, Sharma SK, Bawaskar HS, Kuch U, Chappuis F. Snake bite in South Asia: a review. PLoS Negl Trop Dis. 2010 Jan 26;4(1):e603. pmid:20126271
- 26. Gopalakrishnan M, Yadav P, Mathur R, Midha N, Garg MK. Venom-Induced Consumption Coagulopathy Unresponsive to Antivenom After Echis carinatus sochureki Envenoming. Wilderness Environ Med. 2021 Jun;32(2):221–5. pmid:33781663
- 27. Kumara H, Seneviratne N, Jayaratne DS, Siribaddana S, Isbister GK, Silva A. Severe coagulopathy in Merrem’s hump-nosed pit viper (Hypnale hypnale) envenoming unresponsive to fresh frozen plasma: A case report. Toxicon Off J Int Soc Toxinology. 2019 May;163:19–22. pmid:30885617
- 28. Maduwage K, Scorgie FE, Silva A, Shahmy S, Mohamed F, Abeysinghe C, et al. Hump-nosed pit viper (Hypnale hypnale) envenoming causes mild coagulopathy with incomplete clotting factor consumption. Clin Toxicol Phila Pa. 2013 Aug;51(7):527–31. pmid:23879180
- 29. Ariaratnam CA, Sheriff MHR, Arambepola C, Theakston RDG, Warrell DA. Syndromic approach to treatment of snake bite in Sri Lanka based on results of a prospective national hospital-based survey of patients envenomed by identified snakes. Am J Trop Med Hyg. 2009 Oct;81(4):725–31. pmid:19815895
- 30. Isbister GK. Snake antivenom research: the importance of case definition. Emerg Med J EMJ. 2005 Jun;22(6):399–400. pmid:15911943
- 31.
World Health Organization regional Office for South-East Asia. Guidelines for the management of snakebites. (2nd Edition WHO, New Delhi, 2016) http://apps.searo.who.int/PDS_DOCS/B5255.pdf?ua=1.
- 32. Chaudhari TS, Patil TB, Paithankar MM, Gulhane RV, Patil MB. Predictors of mortality in patients of poisonous snake bite: Experience from a tertiary care hospital in Central India. Int J Crit Illn Inj Sci. 2014 Apr;4(2):101–7. pmid:25024937
- 33. Minghui R, Malecela MN, Cooke E, Abela-Ridder B. WHO’s Snakebite Envenoming Strategy for prevention and control. Lancet Glob Health. 2019 Jul;7(7):e837–8. pmid:31129124
- 34. David S, Matathia S, Christopher S. Mortality predictors of snake bite envenomation in southern India—a ten-year retrospective audit of 533 patients. J Med Toxicol Off J Am Coll Med Toxicol. 2012 Jun;8(2):118–23. pmid:22234395
- 35. Sarkhel S, Ghosh R, Mana K, Gantait K. A hospital based epidemiological study of snakebite in Paschim Medinipur district, West Bengal, India. Toxicol Rep. 2017;4:415–9. pmid:28959667
- 36. Sabitha P, Bammigatti C, Deepanjali S, Suryanarayana BS, Kadhiravan T. Point-of-care infrared thermal imaging for differentiating venomous snakebites from non-venomous and dry bites. PLoS Negl Trop Dis. 2021 Feb;15(2):e0008580. pmid:33600429
- 37. Gopalakrishnan M, Vinod KV, Dutta TK, Shaha KK, Sridhar MG, Saurabh S. Exploring circulatory shock and mortality in viper envenomation: a prospective observational study from India. QJM Mon J Assoc Physicians. 2018 Nov 1;111(11):799–806. pmid:30107433
- 38. Sagar P, Bammigatti C, Kadhiravan T, Harichandrakumar KT, Swaminathan RP, Reddy MM. Comparison of two Anti Snake Venom protocols in hemotoxic snake bite: A randomized trial. J Forensic Leg Med. 2020 Jul;73:101996. pmid:32658754
- 39. Cowley LE, Farewell DM, Maguire S, Kemp AM. Methodological standards for the development and evaluation of clinical prediction rules: a review of the literature. Diagn Progn Res. 2019;3:16. pmid:31463368
- 40. Sullivan LM, Massaro JM, D’Agostino RB. Presentation of multivariate data for clinical use: The Framingham Study risk score functions. Stat Med. 2004 May 30;23(10):1631–60. pmid:15122742
- 41. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. J Clin Epidemiol. 2015 Feb;68(2):134–43. pmid:25579640
- 42. Aujesky D, Fine MJ. The pneumonia severity index: a decade after the initial derivation and validation. Clin Infect Dis Off Publ Infect Dis Soc Am. 2008 Dec 1;47 Suppl 3:S133–139. pmid:18986279
- 43. Hepburn-Brown M, Darvall J, Hammerschlag G. Acute pulmonary embolism: a concise review of diagnosis and management. Intern Med J. 2019 Jan;49(1):15–27. pmid:30324770
- 44. Coura-Vital W, Araújo VEM de, Reis IA, Amancio FF, Reis AB, Carneiro M. Prognostic factors and scoring system for death from visceral leishmaniasis: an historical cohort study in Brazil. PLoS Negl Trop Dis. 2014 Dec;8(12):e3374. pmid:25503575
- 45. Dart RC, Hurlbut KM, Garcia R, Boren J. Validation of a severity score for the assessment of crotalid snakebite. Ann Emerg Med. 1996 Mar;27(3):321–6. pmid:8599491
- 46. Fowler AL, Hughes DW, Muir MT, VanWert EM, Gamboa CD, Myers JG. Resource Utilization After Snakebite Severity Score Implementation into Treatment Algorithm of Crotaline Bite. J Emerg Med. 2017 Dec;53(6):854–61. pmid:29102095
- 47. Kang S, Moon J, Chun B. Does the traditional snakebite severity score correctly classify envenomated patients? Clin Exp Emerg Med. 2016 Mar;3(1):34–40. pmid:27752613
- 48. Padhiyar R, Chavan S, Dhampalwar S, Trivedi T, Moulick N. Snake Bite Envenomation in a Tertiary Care Centre. J Assoc Physicians India. 2018 Mar;66(3):55–9. pmid:30341870
- 49. Nishioka SA. Limitations of the snakebite severity score. Ann Emerg Med. 1996 Sep;28(3):371–2. pmid:8780491
- 50. Jeon YJ, Kim JW, Park S, Shin DW. Risk factor, monitoring, and treatment for snakebite induced coagulopathy: a multicenter retrospective study. Acute Crit Care. 2019 Nov;34(4):269–75. pmid:31743633
- 51. Kim JS, Yang JW, Kim MS, Han ST, Kim BR, Shin MS, et al. Coagulopathy in patients who experience snakebite. Korean J Intern Med. 2008 Jun;23(2):94–9. pmid:18646512
- 52. Wood D, Sartorius B, Hift R. Classifying snakebite in South Africa: Validating a scoring system. South Afr Med J Suid-Afr Tydskr Vir Geneeskd. 2016 Dec 21;107(1):46–51. pmid:28112091
- 53. Saravu K, Somavarapu V, Shastry AB, Kumar R. Clinical profile, species-specific severity grading, and outcome determinants of snake envenomation: An Indian tertiary care hospital-based prospective study. Indian J Crit Care Med Peer-Rev Off Publ Indian Soc Crit Care Med. 2012 Oct;16(4):187–92.
- 54. Kumar V., Maheshwari R., Verma H.K. Toxicity and symptomatic identification of species involved in snakebites in the Indian subcontinent. J. Venom. Anim. Toxins incl. Trop. Dis. 2006;12:3–18.
- 55. Gerardo CJ, Vissoci JRN, de Oliveira LP, Anderson VE, Quackenbush E, Lewis B, et al. The validity, reliability and minimal clinically important difference of the patient specific functional scale in snake envenomation. PloS One. 2019;14(3):e0213077. pmid:30835744
- 56.
World Health Organisation 2018, Factsheet-"Snakebite Envenoming". 2018. Accessed May 20 2018. http://www.who.int/en/news-room/fact-sheets/detail/snakebite-envenoming.
- 57. Saini V, Sardana D, Samra T. Management of snake bite victims in a Tertiary Care Intensive Care Unit in North India. Indian J Crit Care Med Peer-Rev Off Publ Indian Soc Crit Care Med. 2014 Aug;18(8):544–5.
- 58. Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiol Camb Mass. 2010 Jan;21(1):128–38. pmid:20010215