Estimating the burden of leptospirosis in Sri Lanka; a systematic review

We recruited studies and reports containing details of possible cases of human leptospirosis originating from Sri Lanka between 1900 and 2017. All types of publications with primary data were included, and no language restrictions were employed. Both published and unpublished research and reports were included.

For this review, we performed an Internet-based search using Pub Med, MEDLINE®, BIOSIS Previews, Zoological Record, Web of Science Core Collection, Current Contents Connect, KCI-Korean Journal Database, BIOSIS Citation Index, Data Citation Index and SciELO Citation Index. Google Scholar was used to gather all Internet-based literature not indexed in the above databases because most of the local journals are not indexed in the selected databases, and it is common to have technical reports (usually available in Google Scholar) rather than journal articles in the Sri Lankan setup. In addition, we used four bibliographic references to search for local literature: Bibliography of medical publications related to Sri Lanka 1811–1976 [15] and its supplement Bibliography on health in Sri Lanka, 1977–1980 [16] by Peiris and Uragoda; Bibliography of Medical Literature 1980–2005 compiled by the Post Graduate Institute of Medicine (PGIM) Library, Colombo; and the Annotated Bibliography of dissertations and theses Presented to PGIM by Postgraduate Trainees, published by the PGIM. Further, we searched technical reports published by the Medical Research Institute (MRI), papers published on the Ceylon Medical Journal before 2008 (the CMJ website is available after 2008), archived issues of the Sri Lanka Journal of Medical Sciences and the Kandy Medical Journal in four libraries (Sri Lanka Medical Association Library, PGIM library and Colombo and Peradeniya Medical faculty libraries). Throughout the process, we searched reference lists of selected articles to include missing articles from the main search. For national level data, we used several sources. Quarterly epidemiological bulletin (QEB) data of the epidemiology unit of Sri Lanka and the data published by the epidemiology unit in disease trends (http://​www.​epid.​gov.​lk) were compared to obtain the best data set for national “surveillance” data. These surveillance data were from routine notifications from hospitalized patients and outpatients and were considered more “representative” of all cases. The Indoor Morbidity and Mortality Returns (IMMR) available from the Ministry of Health were obtained as “hospital” data. IMMR data are based on diagnoses made during the clinical management at an inward setup. This diagnosis is typically based on clinical criteria and varies widely from physician to physician. IMMR data are limited to hospitalized patients and may represent only moderate to severe cases, as mild patients are treated as outpatients.

The key themes for the search included “Leptospirosis” or “Leptospira” in combination with “Sri Lanka” or “Ceylon”. For the PubMed search, appropriate search terms were identified and translated to MeSH terms where possible. Entry items and keywords used in the initial search related to “Leptospira” are listed in Additional file 2. The PubMed search string used was (“Leptospira”[Title/Abstract]. OR “Leptospirosis”[Title/Abstract]. OR “Weil Disease”[Title/Abstract]. OR “Leptospirosis”[Mesh]. OR “Weil Disease”[Mesh]. OR “Leptospira”[Mesh]) AND (“Ceylon” OR “Sri Lanka”). Other databases were searched through Web of Science, and the search string was TOPIC: (Leptospirosis) OR TOPIC: (Leptospira) OR TOPIC: (Weil’s disease) Refined by: TOPIC: (“Sri Lanka” OR “Ceylon”). The Google Scholar search string used was (“Leptospira”OR “Leptospirosis”OR “Weil’s Disease”) AND (“Ceylon” OR “Sri Lanka”).

After removal of duplicates from different databases, the title and abstract were screened to exclude articles that were not relevant. These included articles that were not directly related to leptospirosis, animal studies and articles without primary data. Conference publications were included if the data were not later published as journal articles or technical papers. Full text reviews were performed for all screen-selected articles to further assess the eligibility criteria. Articles reporting diagnosed cases of leptospirosis and deaths due to leptospirosis were included in the final evidence synthesis. For national disease estimates, we included reports with possible cases of leptospirosis.

The data were extracted using an Excel data sheet by two independent investigators (JW and IK), and the extracted dataset was fully reviewed for confirmation by the SA.

Data items included in this review were used to assess the study and to describe sequelae. These included: study year, authors, duration, study population, sampling procedure, sample size (n), laboratory confirmation, complications and deaths.

Risk of bias in individual studies was assessed using a checklist (Additional file 3). The main risk of bias assessed was due to the study population. First, we searched the methods to see whether measures were taken to include the full study population. Although different study populations were allowed in this review, we assessed whether systematic inclusion of all leptospirosis patients was done either through population screening (population-based studies) or through febrile patient screening. The risk of selection bias was classified as mild, moderate or high based on this assessment. Bias due to misclassification was assessed by examining the diagnostic criteria. If the diagnosis was confirmed using the laboratory criteria given by the leptospirosis burden epidemiology research group (LERG), it was considered a study with a low risk of bias. Studies with positive screening tests but that were not confirmed as “low risk” were considered “moderate risk”, and studies without laboratory confirmation were considered at “high risk” for bias. Most of the patients recruited to the studies were moderate to severe patients. This resulted in an over estimate of the complication and mortality rates, and there were higher numbers of publications from areas where higher number of researchers were conducting studies. Hence, the risk of bias across studies due to “publication bias” or selective reporting is discussed in the results section.

We summarized the data into two main categories: national disease estimates and complications/deaths reported in studies. Construction of 95% confidence limits for observed rates and counts was performed using the Poisson distribution model using the formula \( C\pm {Z}_{1-\propto /2}\sqrt{C} \) for counts and \( R\pm {Z}_{1-\propto /2}\sqrt{\frac{\overline{C}}{n}} \)for rates. Since hospital reported data grossly underestimate the leptospirosis incidence, we added an adjustment factor for disease incidence estimates using published research studies [17]. For this purpose, we searched for studies reporting systematic patient recruitment by screening all febrile patients (low risk for selection bias) together with disaggregated data by type of diagnosis (clinically suspected or not). This was slightly different from the global disease burden estimates that used a robust method of adjustment for underdiagnosis using multicounty data and modelling, which may not be directly applicable for in-country estimation. For Sri Lankan data, the adjustments were done conservatively using the reported under-diagnosed proportion and calculating the 95% confidence intervals for the proportion using the original data. Using these estimated proportions, we calculated high and low estimates for the caseload in each year. For case fatality rate calculations, we used studies with more than 100 confirmed cases (to reduce the chance of error), in which data were collected in defined settings over a period of at least 3 months. We calculated the pooled case fatality rate with 95% CI using the Poisson distribution model.

There were no nationally representative incidence studies, community-based incidence studies or prospective cohort studies that could be used for national disease incidence estimates. Only two passive surveillance databases were available for this purpose: notification data (leptospirosis surveillance data) from the epidemiology unit and hospital admission data from IMMR. Surveillance data grossly underestimated the disease incidence and deaths. From 2004 to 2015, 87,075 patients were registered in IMMR as having leptospirosis, compared to 45,316 reported in the surveillance system. Almost half (n = 41,781, 47.0%) of the hospitalized patients were not included in the surveillance system (Table 1). For deaths, the difference was much larger, with 937 deaths reported during the study period in surveillance data compared to 2600 deaths in hospital data; nearly two thirds of deaths were missing (63.0%). The reported case fatality rate was 2.1% in surveillance data and 3.0% in hospital data.

The reasons for increased incidence after 2008 are not well understood. However, we believe that the increased awareness among treating physicians after the outbreak in 2008 contributed to enhanced reporting in subsequent years.

Data from a single study were available to calculate the hospital underestimation of cases. In the study in Kegalle in 2008, 26 probable cases fulfilling the surveillance criteria were treated as other diseases (hence not included in IMMR), while 175 patients were documented as having leptospirosis during the same period [17]. We used these numbers to estimate the annual caseload with the upper limit of 95% confidence interval for the estimates. After the adjustments, the estimated annual caseload of leptospirosis was 10,423, and the cumulative incidence of leptospirosis that required hospitalization in Sri Lanka during 2008–15 was 52.12 (95% CI 51.69–52.57) per 100,000 people per year (Fig. 2).

Even though leptospirosis was declared a notifiable disease in 1991, major attention was paid to leptospirosis after the heavy outbreak reported in 2008 [18]. Deaths due to leptospirosis were reported in surveillance reports only after this outbreak. However, the number of deaths due to leptospirosis from 2004 to 2008 was available in QEB of the 1st quarter in 2009. Neither surveillance data nor hospital data were based on confirmed cases of leptospirosis but rather were based on suspected patients of leptospirosis. Applying the reported case fatality rate from hospitals to the estimated caseload, the estimated number of annual hospital deaths due to leptospirosis was approximately 311 (95% CI 277–347) (Fig. 2).

Multisystemic complications have been reported due to leptospirosis since the first report in 1959. However, a lack of consistency in reporting complications has made it difficult to estimate the true proportions. The definitions used are neither standard nor complete in most studies. Hence, the classification was done as organ “involvement” rather than specific complications. We calculated the reported rates of complications considering the total confirmed cases as the denominator. Table 2 summarizes the complications reported from Sri Lankan patients.

To estimate the proportion of patients with specific organ involvement, all case reports and case series with specific complications (e.g., case series with pancreatitis) were excluded. The reported rates varied widely among studies with renal involvement (median 48.7%), which was the most common complication followed by liver and cardiac involvement (Table 3).

We identified 22 articles with reported causes of death among patients with leptospirosis. There were 4 articles from two databases, and we combined them for data synthesis. The final set included 5 case reports, 6 case series and 9 cross sectional studies. Table 4 summarizes the reported causes of death mentioned in leptospirosis-related publications in Sri Lanka. The exact cause of death according to ICD 10 was not mentioned in most of the articles. Hence, the causes of death are mentioned using the same wording in the article. Most major complications that can lead to death were reported from in Sri Lanka within the limited literature published since 1959.

No publications assessed the mortality rates of Sri Lanka. However, we recruited studies with data on deaths after assessment for risk of bias. None of the articles reported the true mortality rate, as the sample selection was biased towards the objective of each particular study. Hence, the generalizability of death rates is limited.

Death rates were calculated considering the total number of probable cases of leptospirosis as the denominator. Probable cases of leptospirosis were defined as having positive results in either a screening or confirmatory test, including MAT, ELISA, Culture, PCR or any other serological tests. Table 5 summarizes the reported death rates in Sri Lanka.

The highest death rates were reported from studies performed among Intensive care unit (ICU) patients. At General Hospital Kaluthara, 20 deaths out of 45 ICU patients confirmed for leptospirosis were reported, with a case fatality rate of 44.4% [19]. A study done among ICU patients at National Hospital of Sri Lanka reported a Case Fatality Rate (CFR) of 33.3%. Among usual ward settings, the highest death rate was from Teaching Hospital Peradeniya, with 33 deaths of 227 patients (CFR 14.5%) [20], followed by a study done in 1974 with a CFR of 11.7% [21]. However, in both these studies, there was no information on whether the investigators recruited all cases of leptospirosis through systematic patient enrolment. For CFR estimation, we included only three studies with more than 100 confirmed cases. The estimated pooled CFR was 7.0% (95% CI 5.2–9.4). Based on this CFR, the estimated annual death toll due to leptospirosis in Sri Lanka is approximately 730 (95% CI 542–980).