Can computers measure the chronic disease burden using survey questionnaires? The Symptomatic Diagnosis Study

Abstract

Background Improved information collection systems are critical for more accurately estimating the burden of different chronic conditions around the world. Current estimates are limited by the low amount of medical resources in the developing world and the lack of biometry tests for chronic conditions such as depression or arthritis. Yet, chronic conditions form a substantial part of the global disease burden. Computer-based diagnosis and estimation based on self-reported signs and symptoms ("Symptomatic Diagnosis", or SD) may be a promising method for collecting higher quality information on the chronic disease burden. Methods As part of the Population Health Metrics Research Consortium study, we collected nearly 1,400 questionnaires in Mexico from individuals who suffered from chronic conditions that had been diagnosed with gold standard diagnostic criteria, and individuals who did not suffer from any of the 10 target conditions. We implemented four techniques adopted for verbal autopsy cause-of-death calculation: the Tariff Method, Simplified Symptom Pattern, Random Forest, and King-Lu Direct Estimation. We analyzed the comparative performance between these methods, and compared their performance to current epidemiological measurement techniques for select conditions. Results and discussion The top-performing analytical methods are capable of achieving 68% concordance with true diagnosis, and 0.826 accuracy in their ability to calculate fractions of different causes. SD is also capable of matching or outperforming the performance of current estimation techniques for conditions estimated by questionnaire-based methods. Conclusion Symptomatic Diagnosis is a viable method for producing more detailed estimates of the burden of chronic conditions in areas with low health information infrastructure. This technology can provide myriad benefits to the field of epidemiology, such as higher resolution prevalence data, more flexible data collection, and potentially individual diagnosis for certain conditions.