Ascertaining Cause of Death in Dhulikhel, Nepal: Medical Records and Verbal Autopsy

Abstract

Background: Nepal is experiencing a rapid increase in the burden of non-communicable diseases (NCDs). However, cause of death (COD) data in Nepal is scarce and often unreliable. Health policymakers and planners do not have proper data to inform their funding and programmatic decisions. Objective: We sought to identify the COD distribution and to assess the quality of medical record documentation at Dhulikhel Hospital-Kathmandu University Hospital (DH), a tertiary medical center in suburban Nepal. Additionally, we sought to validate whether the SmartVA application for verbal autopsy could be an effective method for collecting mortality data in Nepal. Methodology: We conducted key informant interviews with staff at DH to develop an understanding of the hospital’s death certification process. Next, we implemented a record review of in-hospital mortality cases. Finally, we used gold standard mortality cases for a pilot study to validate the SmartVA application, which computer certifies verbal autopsies. Results: We identified 135 in-hospital deaths at DH in 2014. We reviewed the medical records and death certificate of 107 cases (26.7% files were missing), including 8 files from 2013. Of the 107 cases reviewed, 44 (41%) were assigned Level 1 and 36 (34%) as Level 2, indicating sufficient evidence for an accurate diagnosis. Of the 66 adult gold standard cases (Level 1 or 2), 15.2% stemmed from infectious causes; 78.8% were caused by cancer, cirrhosis, CVD, COPD, or injury. Top causes of pediatric deaths were preterm delivery (42.9%), pneumonia (14.3%), and birth asphyxia (14.3%). Verbal autopsies were conducted with 48 gold standard cases. We found overall agreement (Kappa) of .46 for adult VAs. A Kappa based on broader ICD-10 categories was .65. Disease specific measures of accuracy varied widely, with sensitivities ranging from 0% (e.g. lung cancer) to 100% (e.g. falls). Cause-Specific Mortality Fraction Accuracy was found to be .58. Discussion: NCDs pose a major threat to suburban Nepalese populations. Ongoing, countrywide mortality data collection will be crucial for evidence-based priority setting. In the hospital, records keeping and documentation was not seen as a major priority. Many files were missing or contained insufficient information for classifying COD. Though not perfect for all causes, we found the SmartVA application to be useful for providing general COD data, specifically in settings where death certification is unavailable. Recommendations: DH has good potential to become an important source of mortality data for Nepal. Because many deaths occur at home in Nepal, verbal autopsy should be implemented in the Dhulikhel Heart Study and scaled up throughout the country as a supplemental method of tracking mortality where death certification is currently unavailable.