The impact of commercial health datasets on medical research and health-care algorithms.

As the health-care industry emerges into a new era of digital health driven by cloud data storage, distributed computing, and machine learning, health-care data have become a premium commodity with value for private and public entities. Current frameworks of health data collection and distribution, whether from industry, academia, or government institutions, are imperfect and do not allow researchers to leverage the full potential of downstream analytical efforts. In this Health Policy paper, we review the current landscape of commercial health data vendors, with special emphasis on the sources of their data, challenges associated with data reproducibility and generalisability, and ethical considerations for data vending. We argue for sustainable approaches to curating open-source health data to enable global populations to be included in the biomedical research community. However, to fully implement these approaches, key stakeholders should come together to make health-care datasets increasingly accessible, inclusive, and representative, while balancing the privacy and rights of individuals whose data are being collected.

Counterintuitive results from observational data: a case study and discussion.

OBJECTIVE: To explore the issue of counterintuitive data via analysis of a representative case in which the data obtained was unexpected and inconsistent with current knowledge. We then discuss the issue of counterintuitive data while developing a framework for approaching such findings. DESIGN: The case study is a retrospective analysis of a cohort of coronary artery bypass graft (CABG) patients. Regression was used to examine the association between perceived pain in the intensive care unit (ICU) and selected outcomes. SETTING: Medical Information Mart for Intensive Care-III, a publicly available, de-identified critical care patient database. PARTICIPANTS: 844 adult patients from the database who underwent CABG surgery and were extubated within 24 hours after ICU admission. OUTCOMES: 30 day mortality, 1 year mortality and hospital length of stay (LOS). RESULTS: Increased pain levels were found to be significantly associated with reduced mortality at 30 days and 1 year, and shorter hospital LOS. A one-point increase in mean pain level was found to be associated with a reduction in the odds of 30 day and 1 year mortality by a factor of 0.457 (95% CI 0.304 to 0.687, p<0.01) and 0.710 (95% CI 0.571 to 0.881, p<0.01) respectively, and a 0.916 (95% CI -1.159 to -0.673, p<0.01) day decrease in hospital LOS. CONCLUSION: The finding of an association between increased pain and improved outcomes was unexpected and clinically counterintuitive. In an increasingly digitised age of medical big data, such results are likely to become more common. The reliability of such counterintuitive results must be carefully examined. We suggest several issues to consider in this analytic process. If the data is determined to be valid, consideration must then be made towards alternative explanations for the counterintuitive results observed. Such results may in fact indicate that current clinical knowledge is incomplete or not have been firmly based on empirical evidence and function to inspire further research into the factors involved.

Feature selection and prediction of treatment failure in tuberculosis.

BACKGROUND: Tuberculosis is a major cause of morbidity and mortality in the developing world. Drug resistance, which is predicted to rise in many countries worldwide, threatens tuberculosis treatment and control. OBJECTIVE: To identify features associated with treatment failure and to predict which patients are at highest risk of treatment failure. METHODS: On a multi-country dataset managed by the National Institute of Allergy and Infectious Diseases we applied various machine learning techniques to identify factors statistically associated with treatment failure and to predict treatment failure based on baseline demographic and clinical characteristics alone. RESULTS: The complete-case analysis database consisted of 587 patients (68% males) with a median (p25-p75) age of 40 (30-51) years. Treatment failure occurred in approximately one fourth of the patients. The features most associated with treatment failure were patterns of drug sensitivity, imaging findings, findings in the microscopy Ziehl-Nielsen stain, education status, and employment status. The most predictive model was forward stepwise selection (AUC: 0.74), although most models performed at or above AUC 0.7. A sensitivity analysis using the 643 original patients filling the missing values with multiple imputation showed similar predictive features and generally increased predictive performance. CONCLUSION: Machine learning can help to identify patients at higher risk of treatment failure. Closer monitoring of these patients may decrease treatment failure rates and prevent emergence of antibiotic resistance. The use of inexpensive basic demographic and clinical features makes this approach attractive in low and middle-income countries.

One-year mortality after recovery from critical illness: A retrospective cohort study.

RATIONALE: Factors associated with one-year mortality after recovery from critical illness are not well understood. Clinicians generally lack information regarding post-hospital discharge outcomes of patients from the intensive care unit, which may be important when counseling patients and families. OBJECTIVE: We sought to determine which factors among patients who survived for at least 30 days post-ICU admission are associated with one-year mortality. METHODS: Single-center, longitudinal retrospective cohort study of all ICU patients admitted to a tertiary-care academic medical center from 2001-2012 who survived ≥30 days from ICU admission. Cox's proportional hazards model was used to identify the variables that are associated with one-year mortality. The primary outcome was one-year mortality. RESULTS: 32,420 patients met the inclusion criteria and were included in the study. Among patients who survived to ≥30 days, 28,583 (88.2%) survived for greater than one year, whereas 3,837 (11.8%) did not. Variables associated with decreased one-year survival include: increased age, malignancy, number of hospital admissions within the prior year, duration of mechanical ventilation and vasoactive agent use, sepsis, history of congestive heart failure, end-stage renal disease, cirrhosis, chronic obstructive pulmonary disease, and the need for renal replacement therapy. Numerous effect modifications between these factors were found. CONCLUSION: Among survivors of critical illness, a significant number survive less than one year. More research is needed to help clinicians accurately identify those patients who, despite surviving their acute illness, are likely to suffer one-year mortality, and thereby to improve the quality of the decisions and care that impact this outcome.

A "datathon" model to support cross-disciplinary collaboration.

In recent years, there has been a growing focus on the unreliability of published biomedical and clinical research. To introduce effective new scientific contributors to the culture of health care, we propose a "datathon" or "hackathon" model in which participants with disparate, but potentially synergistic and complementary, knowledge and skills effectively combine to address questions faced by clinicians. The continuous peer review intrinsically provided by follow-up datathons, which take up prior uncompleted projects, might produce more reliable research, either by providing a different perspective on the study design and methodology or by replication of prior analyses.