Clinical predictors of in-hospital mortality in ... - Wiley Online Library

12 downloads 726 Views 159KB Size Report
Jan 22, 2006 - bite: a retrospective study from a rural hospital in central India ... antivenom, ventilator therapy and renal support systems in patients with snake ...
Tropical Medicine and International Health

doi:10.1111/j.1365-3156.2005.01535.x

volume 11 no 1 pp 22–30 january 2006

Clinical predictors of in-hospital mortality in patients with snake bite: a retrospective study from a rural hospital in central India Shriprakash Kalantri1,2, Amandeep Singh1, Rajnish Joshi1,2, Samuel Malamba2, Christine Ho2, Joseph Ezoua2 and Maureen Morgan2 1 Department of Medicine, Mahatma Gandhi Institute of Medical Sciences, Sevagram, India 2 Division of Epidemiology, University of California at Berkeley, Berkeley CA, USA

Summary

objective To determine the association between selected admission risk factors and in-hospital mortality in patients admitted with venomous snake bite to a rural tertiary care hospital in central India. methods Retrospective cohort study of patients aged 12 years or older admitted to a rural hospital in central India between January 2000 and December 2003 with venomous snake bites. The primary endpoint was in-hospital mortality. We used Cox proportional-hazards regression analysis to evaluate the association between risk factors (home-to-hospital distance, bite-to-hospital time, vomiting, neurotoxicity, urine albumin, serum creatinine concentration and whole-blood clotting time) and in-hospital mortality. results Two hundred and seventy-seven patients [mean age 32 (SD 12) years; 188 men (68%)] were admitted with venomous snake bite, 29 patients (11%) died. The probability of survival at day 7 was 83%. Vomiting [hazard ratio 6.51 (95% CI 1.94–21.77), P £ 0.002], neurotoxicity [hazard ratio 3.15 (95% CI 1.45–6.83), P ¼ 0.004] and admission serum creatinine concentration [hazard ratio 1.35 (95% CI 1.17–1.56), P £ 0.001] were associated with higher risk of death in the adjusted analysis. conclusions In our rural hospital setting, the overall mortality rate was 11 per 100 cases of snake bite. Vomiting, neurotoxicity and serum creatinine are significant predictors of mortality among inpatients with snake bite. These predictors can help clinicians assess prognosis of their patients more accurately and parsimoniously and also serve as useful signposts for clinical decision-making. keywords snake bites, mortality, survival, predictors, risk factors, envenoming, India

Introduction Snake bite, an important cause of death in rural patients in developing countries, is a neglected public health problem. Worldwide, of the estimated 5 million people bitten by snakes each year, about 125 000 die (Murray & Lopez 1996). More than 200 000 cases of snake bite are reported in India each year and 35 000–50 000 of them are fatal. In Maharashtra, a state in India, an estimated 10 000 annual venomous snake bites account for 2000 deaths (Warrell 1999). However, these data are derived from hospitalbased sources, which are likely to grossly underestimate the incidence and mortality of snake bites. About 75% of the Indian population is rural. Most snake bite victims live in villages, seek traditional treatment and many die at home or during transport to hospital (Sharma et al. 2004a). In India, victims of snake bite run a high risk of dying even if they reach hospital. This is 22

because snake venom contains a variety of enzymes and non-enzymatic toxic polypeptides, which affect multiple organs such as kidneys, the coagulation system and respiratory muscles. Most rural hospitals lack the intensive care facilities required for care of patients with multi-organ dysfunction. Also, inappropriate use of antivenom in rural hospitals is common (Warrell 2003). To make more meaningful use of resources such as antivenom, ventilator therapy and renal support systems in patients with snake bite, it is important that the healthcare providers should be able to identify patients with snake bites at high risk of potentially fatal complications. Simple demographic and clinical characteristics could be used to help doctors distinguish between high-risk and low-risk patients. To be useful, the predictors should be simple, accurate and clinically credible. We conducted this retrospective study to evaluate predictors of in-hospital mortality in a rural hospital setting in India.

ª 2006 Blackwell Publishing Ltd

Tropical Medicine and International Health

volume 11 no 1 pp 22–30 january 2006

S. Kalantri et al. Predictors of mortality after snake bite

Materials and methods Setting and study design The Mahatma Gandhi Institute of Medical Sciences (MGIMS), Sevagram is a 648-bed teaching institution in rural central India with 325 000 patient visits and about 5500 patient admissions to medicine wards per year. The hospital, located in a small town in Maharashtra state, serves a predominantly rural population and admits about 100 patients with snake bite each year. All patients with snake bite aged 12 years and older are admitted to the medicine wards of the hospital. They are monitored for at least 24 h and receive lyopholized polyvalent antivenom (Haffkine Bio-pharmaceutical Company Ltd., Bombay, India) if they show progressive venom injury such as local swelling, a clinically important coagulation abnormality [incoaguable blood on 20-minute whole-blood clotting test (20WBCT)] or systemic effects such as ptosis or respiratory weakness. The intensive care unit of the hospital is equipped with a ventilator and has facilities for peritoneal dialysis. Residents and nurses record hospital data on a structured monitoring sheet. Ethical approval for the study was granted by the hospital ethical committee. Data collection We identified all cases of suspected snake bite admitted to the medicine wards of our hospital between January 2000 and December 2003 on the basis of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes for venomous snakes and lizards (E 905.0) and venom (E 989.5). We excluded patients who did not show features of envenoming from the analyses. We used pilot-tested data abstraction forms to collect data on patient demographics and admission clinical variables. The data were checked for consistency and completeness. In the univariate analysis, we regarded the following variables as potential prognostic factors: age, sex, bite-to-hospital time (time taken by the patient to be brought to the hospital after the bite), home-to-hospital distance (distance from the patient’s home to hospital, recorded in km and derived by measuring the radial distance of the patients’ residence from the hospital), diurnal variation (day or night), the site of snake bite, use of tourniquet, local swelling, symptoms (vomiting and neurotoxicity), urine albumin, 20WBCT and serum creatinine concentration. Neurotoxicity was defined as documented ptosis, external ophthalmoloplegia, weakness of neck or bulbar muscles, use of neostigmine or ventilatory support (endotracheal intubation, Ambu bag or a mechanical ventilator). All patients underwent a wholeblood clotting test similar to that described by Warrell

ª 2006 Blackwell Publishing Ltd

et al. (1999). Age, bite-to-hospital interval, home-tohospital distance and serum creatinine concentration were continuous variables; the rest were binary variables (coded as 0¼ absent and 1¼ present). Statistical analysis All analyses were performed using Stata software (Version 8, Stata Corporation, College Station, TX, USA) and R Version 2.01 (R Foundation for Statistical Computing, Vienna, Austria). To compare demographic and clinical characteristics between survivors and non-survivors, we used the t-test for continuous, normally distributed variables; chi-square or Fisher’s exact test, as appropriate for categorical variables and Wilcoxon’s Mann–Whitney U-test for non-parametric variables. All tests were twosided, with a P value of 0.05 or less considered statistically significant. We explicitly considered the time to event for each individual in the study, and analysed the data with ‘survival analysis’ methods. The Kaplan–Meier productlimit estimator was used to estimate survival and for the time-to-event plot. Time to discharge and time to death were investigated with follow-up for all patients starting at hospital admission and ending on day 7. Patients were censored if they were still alive and did not have a poor outcome at the end of follow-up. Event-free subjects were right censored on day 7 after admission to the hospital, because no patient died in the hospital after day 7. The primary end point in this analysis was in-hospital mortality. Mortality was defined as death during the index hospitalization. To identify those predictors with the most significant independent influence on prognosis, we used the log rank test for simple comparisons. Crude hazard ratios were computed to assess the strength of association between risk factors (covariates) and outcome (in-hospital mortality). We used the Cox proportional-hazards regression model for analyses of multiple predictor variables (Cox 1972). This model measures the hazard ratio – the relative effect of a predictive factor on an outcome – by assuming that this relation is constant over time. The assumption of proportional hazards was validated graphically by the log–log time plot. Because many of the risk factors were correlated, collinearity was evaluated by generating correlation matrices and handled by eliminating one of the two collinear variables. Only three variables with a significant unadjusted association with death were included in our final regression model. Since we had only 29 outcome events, this approach accords with accepted statistical practice (Katz 1999). A forward stepwise technique was used in the selection of covariates. For a variable to enter in the model, the P value had to be 0.1. No interactions were entered into the final model because they did not improve fit of the model. Both the crude and the adjusted hazard ratio estimates were computed along with 95% confidence intervals (CI). Results Study patients Between January 1, 2000, and December 31, 2003, a total of 370 snake bite patients, aged 12 years and older, consisting of 242 men and 128 women [mean (SD) age 32 (12) years] were admitted to the hospital. Of these, 93 patients (25%) were excluded from the analysis because they had a non-venomous bite (n ¼ 47), or a dry bite (bite without envenoming; n ¼ 16) or were stung by scorpion and not bitten by a snake (n ¼ 30) (Figure 1). The final dataset comprised records of 277 patients [mean (SD) age 32 (12) years]; [188 (68%) men and 89 (32%) women]. Eighty-four per cent were from rural areas and most patients were bitten engaged in farming activities. Most bites (56%) occurred between June and September, a period of rains and intense farming activity in rural India. Of the 277 patients with venomous snake bites, 81 (29%) patients or bystanders claimed that they could identify the species of the snake on the basis of size, shape, colour or pattern of marking [vipers, 68 (24%), cobras, 11 (4%) and kraits, 2 (1%)]. Table 1 shows the demographic and clinical characteristics of the patients according to their outcome status: survival (n ¼ 248) and death (n ¼ 29). The median biteto-hospital time was 3 h (range: 1–72 h). One-third of patients lived within 10 km of the hospital and 190 patients (69%) had the snake bite between 0600 and 1800 h. All bites were on extremities; legs were three times

Patients with suspected snake bites admitted between January 2000 and December 2003 (n = 370)

Survived (n = 248)

Figure 1 Study profile.

24

Died (n = 29)

Determinants of fatal outcome In the univariate analysis the following risk factors were significantly associated with mortality: home-to-hospital distance, vomiting, neurotoxicity, urine albumin and serum creatinine (Table 2). Age and sex differentials were not significant. In the final model, vomiting [hazard ratio 6.51 (95% CI 1.94–21.77) v2 ¼ 15.34, P ¼ 0.002); neurotoxicity [hazard ratio 3.15 (95% CI 1.45–6.83); v2 ¼ 6.79, P ¼ 0.004) and serum creatinine [hazard ratio 1.35 (95% CI 1.17–1.56); P < 0.001) were independent predictors of mortality. The risk of death was six times higher for those with a history of vomiting and three times higher for those with neurotoxicity compared to patients who did not have these risk factors. After adjustment for other risk factors, for every 1 mg increase in serum creatinine, the likelihood of death increased by a factor of 1.35 (95% CI 1.21–1.55). The Kaplan–Meier estimates showed that patients who did not vomit after snake bite were more likely to survive during the 7-day-study period than those who did [0.98 (95% CI 0.95–1.00) vs. 0.72 (95% CI 0.61–0.84)]. Similarly, compared to those with neurotoxicity, patients without neurotoxicity had better chances of survival [0.89 [95% CI 0.81–0.96) vs. 0.75 (95% CI 0.64–0.87)] (Figure 2). Discussion

Patients with non-venomous bite excluded (n = 93) Patients with venomous bite (n = 277)

more often involved than hands. About 18% received antivenom at the village-based primary health centre before they were referred to our hospital and 37% of patients used at least one of the following first-aid measures: incision of wound, application of tourniquet or both. However, no data were available to quantify the width and pressure of the circumferential ligature or its duration. No patient used a pressure immobilization bandage over the bitten extremity. The common presenting symptoms were pain and swelling at the bite site, vomiting, drooping of eyelids and difficulty breathing. All deaths occurred within 7 days of admission.

Common venomous snakes in central India are cobras (Naja naja), Russell’s viper (Daboia russelii), the saw scaled viper (Echis carinatus) and kraits (Bungarus caeruleus) (Bambery et al. 1993). Vipers, the most abundant snake species in our study area, cause rapid progressive swelling and coagulopathy. Renal failure, unusual after Echis bite, is a classical feature of Russell’s viper envenoming. By contrast, krait and cobra bites are characteristically associated with neurotoxicity. Half the patients in our study presented with local swelling and systemic bleeding and a third presented with neurotoxicity. Clinical

ª 2006 Blackwell Publishing Ltd

Tropical Medicine and International Health

volume 11 no 1 pp 22–30 january 2006

S. Kalantri et al. Predictors of mortality after snake bite

Table 1 Demographic and clinical characteristics of the patients with venomous snake bites

Characteristics

All patients (n ¼ 277)

Survivors (n ¼ 248)

Non-survivors (n ¼ 29)

P value*

Mean age, years (SD) Men (%) Mean bite-to- hospital time, hours (SD) Bite-to-hospital distance >15 km (%) Bite during day (%) Bite on leg (%) Tourniquet use (%) Local swelling (%) Vomiting (%) Neurotoxicity (%) Urine albumin (%) Abnormal whole-blood clotting test (%) Median serum creatinine (mg/dl)  (IQR)

32 188 6.5 85 191 200 29 238 150 88 84 227 1.0

32 165 5.6 81 168 178 25 211 124 71 68 201 1.0

31 23 11.4 4 23 22 4 27 26 17 16 26 1.6

0.481 0.160