This validation study was part of Addis Ababa Mortality Surveillance Program (AAMSP) the then project and in place since 2001 in Addis Ababa, Ethiopia. We validated verbal autopsy data with hospital data of the program during 2007 to 2010.
The sampling frame for this verbal autopsy method; the burial surveillance was conducted in all cemeteries (n ≈ 89) of Addis Ababa since 2001. Since cremation is not practiced in Addis Ababa, burials of deaths are conducted at religious or municipality based cemeteries. In principle thus, the burial surveillance captures all deceased residents of Addis Ababa, although biases exist because residents may die and/or be buried outside the capital just as non-residents may be buried inside Addis Ababa. Some of these biases are mostly identified and corrected while others inevitably go unnoticed.
The surveillance registration was conducted by cemetery based clerks who were regularly trained and supervised. Approximately, 20,000 deaths per year were reported by these cemetery clerks. However, due to financial and logistic reasons; randomly 10% of the deaths were drawn from the burial database for verbal autopsy interview.
We employed retrospective reviews of burials with verbal autopsy technique and retrospective reviews of hospital records in Addis Ababa, Ethiopia.
Data collection for verbal autopsy
Three pairs of field workers who are non health professionals who were trained in the technique visited households of the deceased (minimum one month and maximum three months after the funeral) and selected a respondent who was the person most closely associated with the deceased during the terminal illness. The interview was carried out in Amharic (the national language) once its purpose had been fully explained and consent obtained.
Field workers were recruited locally to ensure a common cultural background with the local community. All had completed secondary school, were experienced in conducting surveys, and had demonstrated the ability to conduct a verbal autopsy interview with insight and empathy. The first several interviews of each field worker were carefully monitored and supervised by field coordinators and researchers. Thereafter, weekly feedback sessions were held on a regular basis, providing an opportunity to appraise the quality of information recorded . The completeness of the VA was approximately 85% where refusal accounts nearly 3%, loss of address 6%, unavailability of the care givers with repeated visit 2% and wrong address 4% which may introduce bias in mortality estimations.
The verbal autopsy questionnaire, adapted from a standardized WHO and INDEPTH Network [17, 18] was translated into Amharic, back translated into English and modified to reflect culturally recognized accepted terms. The questionnaire was divided into four main parts: an open section where the informant freely describes the symptoms and signs preceding death, and their sequence; followed by a closed section in which a basic filtering question, when answered positively, leads to a more detailed enquiry of the particular symptom. Further sections address identification of the caregivers, use of modern and traditional treatments, and lifestyle practices of the deceased.
The physician review method was used to determine causes of death. Three physicians were participating in the review process which were second or third year internal medicine residents of Addis Ababa University recruited to join the university after serving two or more years as a General Practitioner (GP) in any of the public hospitals. We provided them trainings and annual refreshments on the standard verbal autopsy method.
First, each completed questionnaire was reviewed independently by two physicians. If the same diagnosis was reached, this was accepted as the ‘underlying cause of death’, where not, a third physician made a further blind and independent assessment. If two out of three diagnoses corresponded, this would be accepted; otherwise, the three physicians would set for panel, where consensus achieved would be accepted, if not, the cause of death would be described as ‘undetermined’.
In the review process, almost 10% of the verbal autopsy questionnaires required a third physician review for communicable and non communicable major disease categories, but <1% for injury. Less than 1% of the major disease categories required physicians’ panel. Regarding specific causes of deaths, generally <2% of the verbal autopsy questionnaires required third physicians; and <1% required panel. Finally, the research assistant with health background would assign ICD-10 codes according to the international classification of diseases, 10th revision .
Double data entry was done to all the cases, for both the verbal autopsy interview and physician review. Once the data entry was completed, a data manager using STATA driven .do files had been conducting a thorough data cleaning. For the purpose of this paper, we adapted the 2006 Global Burden of Disease classification of causes of death as follows; communicable diseases, non communicable causes, and injuries .
Hospital data collection
A retrospective record review of deaths in 43 public and private hospitals of Addis Ababa from 2007 to 2010 was conducted to validate causes of death reported by verbal autopsy. Nearly, 20,000 adult deaths (15 years and above) of Addis Ababa residents were captured during the study period. Each hospital assigned reviewers who are permanent staffs in the hospitals and centrally three nurses were coordinating, supervising and checking the completeness of the report. Hospital records were assessed by hospital clerks blind to the verbal autopsy diagnosis.
Collected information of validation relevant includes full name, age, sex, and date of death, name of the hospital, and full address of the deceased and the principal cause of death. The data collectors and coordinators had prior relevant experience and provided extensive training on proper review of the medical records and registration books and the use of the data abstraction form.
To capture deaths and complete the relevant information in the hospitals, every attempt was made from patient records and death registry books for patients who died during the study period. In the hospitals diagnosing causes of death was performed by physicians considering patients' history, physical examination, laboratory results and imaging investigations. Only diseases responsible for the death were considered as cause/s of death. Finally, cause of death was coded by nurse coordinators according to the international classification of diseases, 10th revision (ICD-10) . We have listed below specific causes of death with the corresponding ICD-10 assigned.
Cause of Death List: ICD - 10 code; HIV/AIDS: B20–B24; Tuberculosis: A15–A19, B90; Respiratory infections: J00–J06, J10–J18, J20–J22, H65–H66; Meningitis: A39, G00, G03; Malignant neoplasm: C00–C97; Diabetes Mellitus: E10–E14; Cardiovascular diseases: I00–I99; Hypertension: I10–I13; Stroke: I60–I69; Respiratory diseases: J30–J98; Digestive diseases: K20–K92; Chronic Liver Disease: K70, K74; Peptic Ulcer Disease: K25–K27; Genitourinary Diseases: N00–N64, N75–N98; Unintentional injuries: V01–X59, Y40–Y86, Y88, Y89; Road traffic accidents: V01–V04, V06, V09–V80, V87, V89, V99; Intentional injuries (Suicide…): X60–Y09, Y35–Y36, Y870, Y871.
Data were double entered to Access Microsoft office and cleaned using STATA .do files. The 2006 Global Burden of Diseases classification was adapted to classify cause of deaths in our study. This classification categorized diseases into; communicable diseases, non-communicable diseases and all injuries .
Verbal Autopsy interview was conducted after obtaining verbal informed consent from the kin or caregiver of the deceased after explaining the purpose and the procedure of the study. Information sheet prepared in English and translated to local language had been provided. Permission for the study had been also obtained from local authorities. Protocol of the program was approved by Institutional Review Board (IRB) of Medical Faculty, Addis Ababa University, and the Ethiopian Science and Technology Agency. Government and institutional officials, religious leaders at each level had been communicated. Individual information was accessible only to the research team and is kept confidential.
Data management and analysis
The validation consisted of a comparison of verbal autopsy final diagnosis with hospital diagnosis taken as a “gold standard”, followed by calculation of their sensitivity, specificity and positive predictive values. The sensitivity of a verbal autopsy for a particular cause of death such as HIV/AIDS is the proportion of the deceased whose verbal autopsy cause of death is correctly identified as HIV/AIDS out of all those who truly died from HIV/AIDS, while the specificity is the proportion whose cause of death is identified as not HIV/AIDS among those who truly did not die from HIV/AIDS .
Verbal autopsy and hospital data sets were merged using the variables; deceased full name, sex, address, age, place of death and date of death. First we found 1356 deaths occurred in hospitals which were reported with verbal autopsy during 2007 to 2010 period. We merged this verbal autopsy data set with hospital data (n = 20,152, age 15 years and above). Finally, we found 335 deaths for this analysis. This was basically due to the incompleteness of the hospital records and registry books; and differences with age, addresses, and deceased full name, place of death or date of death during merging that might introduce bias. The number of causes of death could be greater than the number of deceased adults in the verbal autopsy and hospital diagnosis since we used two or more causes of death (multiple causes of death) and treated independently.
Sensitivity, specificity and positive predictive value analysis was performed for 3 major categories of diseases; communicable diseases, non communicable diseases and injuries and for each of the major diseases under each category such as HIV/AIDS, tuberculosis, malignant neoplasm, cardiovascular diseases etc. We used chi-squared test to compare proportions between selected verbal autopsy adult deaths, selected hospital adult deaths.
Finally, to show the actual causes of death distribution “actual verbal autopsy diagnoses” and for comparison and completeness of this study, cause specific mortality proportion findings of the double mortality burden study included . The current and the former study were from the same study area, data source and study period.