STATISTICAL METHODS NEEDED IN HEALTH ECONOMICS
Author(s)
Gause D, Gause PSee-It-Analytics, Easton , PA, USA
OBJECTIVE: To review important statistical methods used in health economics. BACKGROUND: During two decades in health economics at Novartis Pharmaceuticals, statistician Doug Gause regularly presented friendly one-page illustrations of statistical concepts in health economics. Thirty of these “Analysis Notes” are presented on this poster, along with a magnifying glass for easier viewing. ANALYSIS NOTES: Making inferences: Confidence interval estimation instead of hypothesis testing, and bootstrapping to avoid assumptions. Analyzing costs: Two-part GLM models instead of transformations. Measuring compliance: Avoiding adherence bias in calculating MPR. Competing risks of discontinuing, adding, switching, or remaining on drug. Data visualization: Regression trees, drug-o-grams, mosaic plots, parallel coordinate plots, and sequence plots to visualize patterns. ( Lets reserve pie charts for the Baker's Convention ! ) Covariate adjustment: Marginal means/IPTW, confounders vs. colliders, instrument variables, and time dependent covariates to incorporate changes over time. Historical data: Bayesian analysis for bias adjusting, and meta-analysis to incorporate results from multiple studies. Avoiding pitfalls: Be aware of the “individualized” fallacy and the “ecological“ fallacy. Differential follow-up: Person-time and survival analysis to accommodate censoring. Predictive modeling: Nearest neighbor vs. recursive partitioning vs. neural networks. Project planning: Sample size to handling multiplicity. Clustered/longitudinal data: Random effects models to account for correlated observations from the same physician or repeated measurements from the same patient. Cost-effectiveness: Angles instead of ratios, and incremental net benefit to handle multiple effectiveness measures. Heterogeneous data: Quantile and polytomous regression to compare different MPR distributions. CONCLUSION: Statistics needs to be taught in small friendly doses tied to practical problems. Each analysis note presented by Doug Gause has sketches drawn by artist Pat Gause.
Conference/Value in Health Info
2010-05, ISPOR 2010, Atlanta, GA, USA
Value in Health, Vol. 13, No. 3 (May 2010)
Code
PMC47
Topic
Clinical Outcomes
Topic Subcategory
Relating Intermediate to Long-term Outcomes
Disease
Multiple Diseases