Introduction

Globally, an estimated 500,000 new cases of visceral leishmaniasis (VL) or kala-azar, a ‘Neglected Tropical Disease’, occur annually [1]. More than 90% of the global burden of visceral leishmaniasis (VL) is contributed by six countries: Bangladesh, Brazil, Ethiopia, India, South Sudan and Sudan [2]. The parasite Leishmania donovani is the main causative agent of VL in India, Nepal, and Bangladesh, where it is transmitted by the sand fly vector Phlebotomus argentipes [3]. In Indian subcontinent, as per the prevailing knowledge of transmission dynamics, humans are considered to be the sole carrier of the disease (anthroponotic) and transmission pattern is believed to be peridomestic i.e. transmitted by the bite of female sand flies in and around the home [4, 5].

Based on recent estimates, in India, 70%–90% VL cases occur in a single state–Bihar [6, 7]. The disease was considered to be at the brink of elimination in Bihar during 1960s following widespread DDT use under National Malaria Eradication Program [8, 9]. However, in absence of a dedicated VL elimination strategy and surveillance system, the disease saw a resurgence a decade later [9, 10]. Since then VL has been widely acknowledged as a major public health concern in India. Still a significant proportion of VL caseload in Bihar was assumed to remain outside the public health detection system until recent past [11, 12]. Absence of a robust surveillance system as well as simple and inexpensive diagnostic tests were identified as potential reasons for the poor tracking of VL cases in Indian subcontinent [11, 13].

In order for an effective kala-azar elimination program in Bihar, identification of high endemic foci is essential. Health Management Information System (HMIS), the current source of population level information on VL, has been reported to suffer from various methodological and operational problems including underreporting and ill-defined population frame [11, 12]. For precise identification of blocks having VL incidence above the elimination level (i.e. annual incidence of 1/10000 population or more), a situation assessment project was undertaken in January’2013 [7]. It was also expected that this assessment would help in evaluating the effectiveness of various elimination strategies (e.g. treatment by amphotericin-B or miltefosine) and inform the kala-azar elimination program in India on the incidence of VL associated mortality and its epidemiologic correlates.

In absence of early diagnosis and treatment, fatality was detected to be very high among symptomatic VL cases [14, 15]. Even among treatment-recipients, the measured probability of dying within two years of the onset of symptoms were quite high [7, 16]. Further, among many diagnosed patients, outcome remained unknown due to lack of systematic follow up and cause of death remained obscured among many suspected cases in the community. Thus, alike incidence, it can be hypothesized that mortality due to VL also remain grossly under-reported. To the best of our knowledge, no large community-based study did yet report on either the incidence of mortality or its antecedents among symptomatic VL cases in India. The current exercise, therefore, was undertaken to address this knowledge gap. Our objectives were to provide a reliable estimate of symptomatic VL case load and mortality in selected districts of Bihar, and to identify the epidemiologic and treatment-related predictors of VL mortality.

Methods

Results

Through the index case tracing method, 5770 listed VL cases were identified from the public facilities. Among them, 4962 cases (or their next-of-kin for those who were dead) were successfully traced to their residence. Additionally, records of 1119 cases meeting the study eligibility criteria were obtained from private facilities and through snowballing technique. Following removal of duplicate listings, total 5432 eligible cases were thus recruited. Further, for the current analysis, we did not include i) cases with post-kala-azar dermal leishmaniasis, ii) cases with ambiguous mortality data, and iii) cases for which reported onset of symptoms occurred more than two years before date of diagnosis (because of reliability of recall). Thus, another 507 cases were excluded from the final dataset. Therefore, information on 4925 VL cases who were diagnosed between January 2012 and June 2013 was used for the current exercise. The distribution of cases identified through different sources and at various stages of the study is presented in supporting information S2 Table.

The mean age of the VL patients was 24.9 years (SD ±17.6 years) and about 59% of them were males. More than 80% belonged to either marginalized or semi-marginalized castes. About two-thirds (67%) of the eligible respondents lived in mud/thatched (Kaccha) houses. Majority of the cases not only got tested for VL at a public laboratory (63%) but also received treatment exclusively from public facilities (78%). About 5% patients did not complete their course of treatment, with proportion of treatment defaulter being higher among those prescribed SSG (8%) compared to miltefosine (4%) or amphotericin-B (4%). About 3% VL patients died during the interval between diagnosis and case tracing. The incidence rate of death among symptomatic VL patients was 3.2/100 person-years. The socio-demographic and disease-related characteristics of VL patients are depicted in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Comparison of socio-demographic and disease-related characteristics between the dead and surviving (till date of interview) VL cases in 8 districts of Bihar. 2012–13 (N = 4925)

https://doi.org/10.1371/journal.pntd.0005150.t001

Bivariate comparison of the characteristics between the dead and the alive (till the date of interview) VL patients revealed that there were significant differences in age (higher proportion of older patients among the dead) and caste (higher proportion of marginalized caste among the dead) distribution between the dead and the alive (Table 1). Moreover, patients who completed the course of their therapy, received earlier diagnosis of VL and those who received treatment from the public facilities were less likely to die. Regarding Kaplan-Meier estimates of the probability of survival in the period following diagnosis (Fig 1), there were statistically significant differences (log rank tests) in cumulative survival probabilities across the categories of age, caste, treatment facility, diagnosis interval and treatment completion.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

Estimates of the cumulative survival probabilities (Kaplan-Meier estimates) of VL patients, from the time of diagnosis, stratified by the categories of: (A) age of the patient; (B) gender of the patient; (C) caste of the patient; (D) place from which VL treatment was received; (E) facility where VL was diagnosed; (F) interval between symptom onset and diagnosis; (G) type of drug used for VL treatment; and (H) completion of the course of VL treatment. Bihar, 2012–13. (N = 4925).

https://doi.org/10.1371/journal.pntd.0005150.g001

Table 2 depicts the hazard ratios (and 95% confidence intervals) of death among VL patients associated with various predictors. Those who received treatment exclusively from public facilities were found to have lower risks of death compared to those treated (completely or partially) at private facilities [Unadjusted hazard ratio (HR) = 0.52; 95% confidence interval (CI) = 0.37, 0.72; Adjusted: HR = 0.61; 95% CI = 0.43, 0.86]. Getting an early diagnosis (within 30 days of onset of symptoms) was associated with lower hazard of death, compared to those diagnosed after 30 days (Unadjusted: HR = 0.67; 95% CI = 0.48–0.93; Adjusted: HR = 0.66; 95% CI = 0.47–0.92). There were no significant differences in terms of the drugs used for VL treatment. Irrespective of the drug, patients who completed treatment course had 97% lower risk of death (Unadjusted: HR = 0.03; 95% CI = 0.02, 0.04; Adjusted: HR = 0.03; 95% CI = 0.02, 0.05), compared to those who did not complete treatment. Tests for proportional hazards assumption (Schoenfeld residuals test), revealed that none of the CPH models had a significant p-value for the test of non-zero slope. Thus, the proportional hazards assumption could not be rejected for any of the CPH models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Hazard ratios (and 95% confidence intervals) of death among VL cases, since the time of diagnosis, from crude and adjusted Cox proportional hazards models.

Bihar, 2012–13. (N = 4925).^*

https://doi.org/10.1371/journal.pntd.0005150.t002

Discussion

In the current article, we present the findings of the VL situation assessment initiative aimed at estimating the burden and mortalities associated with symptomatic VL in the Indian state of Bihar. Previous estimates of VL incidence in Bihar, which has the highest burden of VL in India, were based either on the data from public facilities or surveys conducted in limited geographical areas with high endemic foci [11, 12]. Moreover, there existed very few studies on VL-associated mortalities in India. From the perspective of India’s kala-azar elimination program, which was principally focused on prevention, understanding the correlates of VL-associated deaths could help in designing interventions targeted at reduction of VL mortalities and prioritizing deliveries of essential health services to VL patients.

The case fatality of symptomatic VL was found to be around 3%, lower than the value reported by a large scale VL mortality study conducted in Bangladesh [13]. The comparatively higher proportion of death reported in the Bangladesh study could be attributed to the large tribal ethnic population included in that study, with about 83% VL deaths occurring among tribals [13]. Moreover, unlike Bihar, about one-third of the death cases in Bangladesh did not receive any treatment. Thus, the lower mortality seen in Bihar, compared to Bangladesh, might be a consequence of wider access to modern treatment for VL.

The proportion of males were higher among VL cases. This is similar to the findings from previous studies in the same geographical region and could be attributed to their sleeping habits (sleeping outside the house) and occupational exposures (e.g. farming) [6]. Also, families of about 61% symptomatic VL patients were owners of cattle. This was not surprising given that cattle-sheds are known breeding site for the vector in Bihar [17]. For majority of patients, the interval between the onset of symptoms and diagnosis was more than 30 days. An earlier study conducted in Bihar and neighboring Nepal also reported on the issue of diagnostic delay in Bihar [18]. However, unlike the said study, most of the VL patients included in the current research sought treatment from Government-run facilities. Minimizing the delay in diagnosis and treatment, besides being prognostically beneficial for individual patients, is crucial for restricting the disease transmission as the infected hosts serve as source of leishmania amastigotes for sandflies. About 8% patients were prescribed SSG, despite the widely reported resistance against it [19]. Moreover, the likelihood of non-completion of treatment was much higher among those prescribed SSG, possibly due to adverse effects or prolonged parenteral therapy or both [20].

Probability of death was higher among people from marginalized castes corroborating the observation in Bangladesh [13]. As marginalized people in India belong mostly to lower socio-economic strata, the observed higher mortality could be an outcome of their low awareness levels and/or poorer access to healthcare [17, 21]. The impact of economic status on mortality was also evident from the fact that ownership of cattle, which is a surrogate for better economic status in rural Bihar, was found to be associated with lower mortality risk. This is not surprising as economic status is a principal determinant of access to healthcare and, in turn, disease outcome. Moreover, ownership of cattle might have been associated with milk consumption and better nutritional status, which could have potentially lowered the risk of mortality [22, 23]. Further, treatment at public facilities was found to be associated with lower risk of death compared to getting treated by private healthcare providers. In rural Bihar’s scenario, unqualified practitioners are often the mainstay of health services. Despite the fact that health services provided at public health facilities are free and there are concerns about the quality of care provided by the rural medical practitioners, residents of hard-to-reach areas often have to rely on such practitioners. Poor access to quality healthcare among VL patients in India and other resource-limited settings have long been recognized as a major concern [24]. Therefore, future policies aimed at reducing VL mortalities in India should address issues related to proper healthcare access among residents of rural areas, especially those belonging to the marginalized communities.

Proportional hazards analysis revealed that diagnosis within 30 days of symptom onset and completion of treatment course to be significant predictors of survival among symptomatic VL cases. Earlier diagnosis is likely to be associated with earlier treatment as well–which is a widely recognized determinant of survival[25]. Treatment completion, irrespective of the prescribed drug, with 96%-97% reduced risk of mortality, emerged as a strong predictor of survival. Treatment adherence, besides being essential for cure, is also important for preventing emergence of drug resistance. Therefore, targeted interventions aimed at strengthening the adherence improvement measures, such as rigorous counselling and treatment follow-up—especially for potential defaulters, need to be emphasized upon.

There were some important limitations of the present undertaking. First, the current study included only the symptomatic VL cases. However, the burden of asymptomatic VL has been reported to be much larger compared to the symptomatic cases [26]. Nevertheless, as mortality due to VL is expected to occur mostly among symptomatic cases, the mortality estimate (and its predictors) is presumed to be somewhat accurate. Second, some of the dates used in the analysis (especially date of onset of illness), an essential component of survival analysis, were approximate and based on recall. Therefore, the possibility of some amount of information bias in the estimates should be borne in mind. Third, because of unavailability of past treatment documents, it was not possible for us to determine whether a VL case was new or relapse and if there were any associated comorbidities (e.g. HIV). Another potential limitation of our findings is that our analyses could not account for all possible factors that might have contributed to mortality among VL patients (nutritional status, comorbidities etc.), as the required information were not collected. Finally, it was not possible to ascertain the exact causes of death for all cases (where no medical documents were available). However, such missing information were likely to be non-differentially distributed across the categories of the variables incorporated in the statistical model(s). Because of the aforementioned limitations, caution should be exercised while interpreting the results of our analyses (and extrapolating the findings to the national level).

Notwithstanding the limitations, the findings of this first-ever attempt at identifying the epidemiologic and health-system related correlates of VL-associated mortalities in Bihar probably could generate important insights. It was revealed that the interventions aimed at reducing mortalities should specifically target the marginalized communities and focus on improving access to early diagnostic and quality treatment services along with strengthening the treatment adherence measures. We expect our findings to contribute towards the efforts of reducing VL mortalities and help in epidemiologic transition of VL to a non-fatal low-endemic disease in Bihar.

Supporting Information