Constructing synthetic populations in the age of big data.

BACKGROUND: To develop public health intervention models using micro-simulations, extensive personal information about inhabitants is needed, such as socio-demographic, economic and health figures. Confidentiality is an essential characteristic of such data, while the data should reflect realistic scenarios. Collection of such data is possible only in secured environments and not directly available for open-source micro-simulation models. The aim of this paper is to illustrate a method of construction of synthetic data by predicting individual features through models based on confidential data on health and socio-economic determinants of the entire Dutch population. METHODS: Administrative records and health registry data were linked to socio-economic characteristics and self-reported lifestyle factors. For the entire Dutch population (n = 16,778,708), all socio-demographic information except lifestyle factors was available. Lifestyle factors were available from the 2012 Dutch Health Monitor (n = 370,835). Regression model was used to sequentially predict individual features. RESULTS: The synthetic population resembles the original confidential population. Features predicted in the first stages of the sequential procedure are virtually similar to those in the original population, while those predicted in later stages of the sequential procedure carry the accumulation of limitations furthered by data quality and previously modelled features. CONCLUSIONS: By combining socio-demographic, economic, health and lifestyle related data at individual level on a large scale, our method provides us with a powerful tool to construct a synthetic population of good quality and with no confidentiality issues.

Mapping chronic disease prevalence based on medication use and socio-demographic variables: an application of LASSO on administrative data sources in healthcare in the Netherlands.

BACKGROUND: Policymakers generally lack sufficiently detailed health information to develop localized health policy plans. Chronic disease prevalence mapping is difficult as accurate direct sources are often lacking. Improvement is possible by adding extra information such as medication use and demographic information to identify disease. The aim of the current study was to obtain small geographic area prevalence estimates for four common chronic diseases by modelling based on medication use and socio-economic variables and next to investigate regional patterns of disease. METHODS: Administrative hospital records and general practitioner registry data were linked to medication use and socio-economic characteristics. The training set (n = 707,021) contained GP diagnosis and/or hospital admission diagnosis as the standard for disease prevalence. For the entire Dutch population (n = 16,777,888), all information except GP diagnosis and hospital admission was available. LASSO regression models for binary outcomes were used to select variables strongly associated with disease. Dutch municipality (non-)standardized prevalence estimates for stroke, CHD, COPD and diabetes were then based on averages of predicted probabilities for each individual inhabitant. RESULTS: Adding medication use data as a predictor substantially improved model performance. Estimates at the municipality level performed best for diabetes with a weighted percentage error (WPE) of 6.8%, and worst for COPD (WPE 14.5%)Disease prevalence showed clear regional patterns, also after standardization for age. CONCLUSION: Adding medication use as an indicator of disease prevalence next to socio-economic variables substantially improved estimates at the municipality level. The resulting individual disease probabilities could be aggregated into any desired regional level and provide a useful tool to identify regional patterns and inform local policy.

Estimating disease prevalence from drug utilization data using the Random Forest algorithm.

BACKGROUND: Aggregated claims data on medication are often used as a proxy for the prevalence of diseases, especially chronic diseases. However, linkage between medication and diagnosis tend to be theory based and not very precise. Modelling disease probability at an individual level using individual level data may yield more accurate results. METHODS: Individual probabilities of having a certain chronic disease were estimated using the Random Forest (RF) algorithm. A training set was created from a general practitioners database of 276 723 cases that included diagnosis and claims data on medication. Model performance for 29 chronic diseases was evaluated using Receiver-Operator Curves, by measuring the Area Under the Curve (AUC). RESULTS: The diseases for which model performance was best were Parkinson's disease (AUC = .89, 95% CI = .77-1.00), diabetes (AUC = .87, 95% CI = .85-.90), osteoporosis (AUC = .87, 95% CI = .81-.92) and heart failure (AUC = .81, 95% CI = .74-.88). Five other diseases had an AUC >.75: asthma, chronic enteritis, COPD, epilepsy and HIV/AIDS. For 16 of 17 diseases tested, the medication categories used in theory-based algorithms were also identified by our method, however the RF models included a broader range of medications as important predictors. CONCLUSION: Data on medication use can be a useful predictor when estimating the prevalence of several chronic diseases. To improve the estimates, for a broader range of chronic diseases, research should use better training data, include more details concerning dosages and duration of prescriptions, and add related predictors like hospitalizations.

Potential gains in health expectancy by improving lifestyle: an application for European regions.

BACKGROUND: Prevention aiming at smoking, alcohol consumption, and BMI could potentially bring large gains in life expectancy (LE) and health expectancy measures such as Healthy Life Years (HLY) and Life Expectancy in Good Perceived Health (LEGPH) in the European Union. However, the potential gains might differ by region. METHODS: A Sullivan life table model was applied for 27 European countries to calculate the impact of alternative scenarios of lifestyle behavior on life and health expectancy. Results were then pooled over countries to present the potential gains in HLY and LEGPH for four European regions. RESULTS: Simulations show that up to 4 years of extra health expectancy can be gained by getting all countries to the healthiest levels of lifestyle observed in EU countries. This is more than the 2 years to be gained in life expectancy. Generally, Eastern Europe has the lowest LE, HLY, and LEGPH. Even though the largest gains in LEPGH and HLY can also be made in Eastern Europe, the gap in LE, HLY, and LEGPH can only in a small part be closed by changing smoking, alcohol consumption, and BMI. CONCLUSION: Based on the current data, up to 4 years of good health could be gained by adopting lifestyle as seen in the best-performing countries. Only a part of the lagging health expectancy of Eastern Europe can potentially be solved by improvements in lifestyle involving smoking and BMI. Before it is definitely concluded that lifestyle policy for alcohol use is of relatively little importance compared to smoking or BMI, as our findings suggest, better data should be gathered in all European countries concerning alcohol use and the odds ratios of overconsumption of alcohol.