Naive Bayes classifiers for verbal autopsies: comparison to physician-based classification for 21,000 child and adult deaths.

BACKGROUND: Verbal autopsies (VA) are increasingly used in low- and middle-income countries where most causes of death (COD) occur at home without medical attention, and home deaths differ substantially from hospital deaths. Hence, there is no plausible "standard" against which VAs for home deaths may be validated. Previous studies have shown contradictory performance of automated methods compared to physician-based classification of CODs. We sought to compare the performance of the classic naive Bayes classifier (NBC) versus existing automated classifiers, using physician-based classification as the reference. METHODS: We compared the performance of NBC, an open-source Tariff Method (OTM), and InterVA-4 on three datasets covering about 21,000 child and adult deaths: the ongoing Million Death Study in India, and health and demographic surveillance sites in Agincourt, South Africa and Matlab, Bangladesh. We applied several training and testing splits of the data to quantify the sensitivity and specificity compared to physician coding for individual CODs and to test the cause-specific mortality fractions at the population level. RESULTS: The NBC achieved comparable sensitivity (median 0.51, range 0.48-0.58) to OTM (median 0.50, range 0.41-0.51), with InterVA-4 having lower sensitivity (median 0.43, range 0.36-0.47) in all three datasets, across all CODs. Consistency of CODs was comparable for NBC and InterVA-4 but lower for OTM. NBC and OTM achieved better performance when using a local rather than a non-local training dataset. At the population level, NBC scored the highest cause-specific mortality fraction accuracy across the datasets (median 0.88, range 0.87-0.93), followed by InterVA-4 (median 0.66, range 0.62-0.73) and OTM (median 0.57, range 0.42-0.58). CONCLUSIONS: NBC outperforms current similar COD classifiers at the population level. Nevertheless, no current automated classifier adequately replicates physician classification for individual CODs. There is a need for further research on automated classifiers using local training and test data in diverse settings prior to recommending any replacement of physician-based classification of verbal autopsies.

Performance of four computer-coded verbal autopsy methods for cause of death assignment compared with physician coding on 24,000 deaths in low- and middle-income countries.

BACKGROUND: Physician-coded verbal autopsy (PCVA) is the most widely used method to determine causes of death (CODs) in countries where medical certification of death is uncommon. Computer-coded verbal autopsy (CCVA) methods have been proposed as a faster and cheaper alternative to PCVA, though they have not been widely compared to PCVA or to each other. METHODS: We compared the performance of open-source random forest, open-source tariff method, InterVA-4, and the King-Lu method to PCVA on five datasets comprising over 24,000 verbal autopsies from low- and middle-income countries. Metrics to assess performance were positive predictive value and partial chance-corrected concordance at the individual level, and cause-specific mortality fraction accuracy and cause-specific mortality fraction error at the population level. RESULTS: The positive predictive value for the most probable COD predicted by the four CCVA methods averaged about 43% to 44% across the datasets. The average positive predictive value improved for the top three most probable CODs, with greater improvements for open-source random forest (69%) and open-source tariff method (68%) than for InterVA-4 (62%). The average partial chance-corrected concordance for the most probable COD predicted by the open-source random forest, open-source tariff method and InterVA-4 were 41%, 40% and 41%, respectively, with better results for the top three most probable CODs. Performance generally improved with larger datasets. At the population level, the King-Lu method had the highest average cause-specific mortality fraction accuracy across all five datasets (91%), followed by InterVA-4 (72% across three datasets), open-source random forest (71%) and open-source tariff method (54%). CONCLUSIONS: On an individual level, no single method was able to replicate the physician assignment of COD more than about half the time. At the population level, the King-Lu method was the best method to estimate cause-specific mortality fractions, though it does not assign individual CODs. Future testing should focus on combining different computer-coded verbal autopsy tools, paired with PCVA strengths. This includes using open-source tools applied to larger and varied datasets (especially those including a random sample of deaths drawn from the population), so as to establish the performance for age- and sex-specific CODs.

Performance criteria for verbal autopsy-based systems to estimate national causes of death: development and application to the Indian Million Death Study.

BACKGROUND: Verbal autopsy (VA) has been proposed to determine the cause of death (COD) distributions in settings where most deaths occur without medical attention or certification. We develop performance criteria for VA-based COD systems and apply these to the Registrar General of India's ongoing, nationally-representative Indian Million Death Study (MDS). METHODS: Performance criteria include a low ill-defined proportion of deaths before old age; reproducibility, including consistency of COD distributions with independent resampling; differences in COD distribution of hospital, home, urban or rural deaths; age-, sex- and time-specific plausibility of specific diseases; stability and repeatability of dual physician coding; and the ability of the mortality classification system to capture a wide range of conditions. RESULTS: The introduction of the MDS in India reduced the proportion of ill-defined deaths before age 70 years from 13% to 4%. The cause-specific mortality fractions (CSMFs) at ages 5 to 69 years for independently resampled deaths and the MDS were very similar across 19 disease categories. By contrast, CSMFs at these ages differed between hospital and home deaths and between urban and rural deaths. Thus, reliance mostly on urban or hospital data can distort national estimates of CODs. Age-, sex- and time-specific patterns for various diseases were plausible. Initial physician agreement on COD occurred about two-thirds of the time. The MDS COD classification system was able to capture more eligible records than alternative classification systems. By these metrics, the Indian MDS performs well for deaths prior to age 70 years. The key implication for low- and middle-income countries where medical certification of death remains uncommon is to implement COD surveys that randomly sample all deaths, use simple but high-quality field work with built-in resampling, and use electronic rather than paper systems to expedite field work and coding. CONCLUSIONS: Simple criteria can evaluate the performance of VA-based COD systems. Despite the misclassification of VA, the MDS demonstrates that national surveys of CODs using VA are an order of magnitude better than the limited COD data previously available.

Comparison of physician-certified verbal autopsy with computer-coded verbal autopsy for cause of death assignment in hospitalized patients in low- and middle-income countries: systematic review.

BACKGROUND: Computer-coded verbal autopsy (CCVA) methods to assign causes of death (CODs) for medically unattended deaths have been proposed as an alternative to physician-certified verbal autopsy (PCVA). We conducted a systematic review of 19 published comparison studies (from 684 evaluated), most of which used hospital-based deaths as the reference standard. We assessed the performance of PCVA and five CCVA methods: Random Forest, Tariff, InterVA, King-Lu, and Simplified Symptom Pattern. METHODS: The reviewed studies assessed methods' performance through various metrics: sensitivity, specificity, and chance-corrected concordance for coding individual deaths, and cause-specific mortality fraction (CSMF) error and CSMF accuracy at the population level. These results were summarized into means, medians, and ranges. RESULTS: The 19 studies ranged from 200 to 50,000 deaths per study (total over 116,000 deaths). Sensitivity of PCVA versus hospital-assigned COD varied widely by cause, but showed consistently high specificity. PCVA and CCVA methods had an overall chance-corrected concordance of about 50% or lower, across all ages and CODs. At the population level, the relative CSMF error between PCVA and hospital-based deaths indicated good performance for most CODs. Random Forest had the best CSMF accuracy performance, followed closely by PCVA and the other CCVA methods, but with lower values for InterVA-3. CONCLUSIONS: There is no single best-performing coding method for verbal autopsies across various studies and metrics. There is little current justification for CCVA to replace PCVA, particularly as physician diagnosis remains the worldwide standard for clinical diagnosis on live patients. Further assessments and large accessible datasets on which to train and test combinations of methods are required, particularly for rural deaths without medical attention.