Research methods for Pharmaceutical Policy Evaluation 藥物政策評估研究方法

Transcription

1 Pre-conference course - Introduction to Drug Utilization Research Research methods for Pharmaceutical Policy Evaluation 藥物政策評估研究方法 徐之昇 Jason Hsu, Ph.D. Assistant Professor Institute of Clinical Pharmacy and Pharmaceutical Sciences National Cheng Kung University, Taiwan 國立成功大學臨床藥學與藥物科技研究所助理教授

2 Disclosures There is no potential conflict of interest relevant to this presentation. Some slides in this presentation are adopted from a prior ICPE short course on drug utilization by Colin Dormuth ICPE 2

3 Contents 1. Introduction of Pharmaceutical Policy Research 2. Interrupted Time Series Design and Segmented Regression for Pharmaceutical Policy Evaluation 3. Examples

4 Scope of Pharmaceutical Policy Analysis Preclinical Phase Clinical Phase Post-marketing FDA review process (IND, NDA) Pharmaceutical Development and Health Service Manufacturing (Quality) Effectiveness / Risk Pharmaceutical Care and Rational Drug Use Drug Price and Reimbursement

5 Pharmaceutical Policy Cycle Retrospective Analysis Prospective Analysis Evaluation Policy Performance Forecasting Find Q Solve Q Policy Outcomes Policy Problems Policy Futures (Design) Monitoring Policy Actions Recommendation Dunn, William N. (2008) Public Policy Analysis: An Introduction. 4 th Edition Eengland Cliffs, NJ: Pprentice-Hall.

6 Discipline of Pharmaceutical Policy Analysis Pharmaceutical Policy and Regulatory Science Medical Decision Making Science Development of Pharmaceutical Industry Social Pharmacy: Pharmacology, Clinical Pharmacy, Pharmacoepidemiology, Pharmacoeconomics, Political Science, Management.

7 Pharmaceutical Policy Analytic Methods 1. Problem Structuring Medicine + Social Science 1. Policy Forecasting Statistics (Prospective Analysis) (Time Series) 2. Policy Design Economics (CEA, RBA, GT ) 3. Policy Recommendation Marketing 4. Policy Monitoring Management + Statistics 5. Policy Evaluation Statistics (Retrospective Analysis) (ITS)

8 Contents 1. Introduction of Pharmaceutical Policy Research 2. Interrupted Time Series Design and Segmented Regression for Pharmaceutical Policy Evaluation 3. Examples

9 Classification of Quantitative Analysis Analysis Data Type Data Source Objective Method Special Functions Cross sectional Static (one point time) 1 st, 2 nd before, nowadays Multivariate Statistical Analysis Longitudinal Dynamic (time series) 1 st, 2 nd before, nowadays and future Time Series Analysis Assessing the impact of interventions

10 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time

11 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy? Time

12 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time

13 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy?? Time

14 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time

15 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time

16 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time

17 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time 2005

18 Why Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time 2005

19 Threats to Internal Validity - When an observed effect might be due to: History An event occurring between pre- and post-intervention when the event is not the intervention of research interest Maturation Subjects growing older, wiser, healthier (or sicker), etc. between pre and post when this maturation is not the intervention of interest Instrumentation A change in the measuring instrument (either humans or test) Source: Cook & Campbell, 1979

20 Threats to Internal Validity - When an observed effect might be due to: Selection Pre-intervention differences between people in one experimental group vs. another Interactions with Selection Selection - history different local histories Selection - maturation maturing at different speeds Selection - instrumentation ceiling effects on one measure Non-equivalent Control Group Design Source: Cook & Campbell, 1979

21 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Predicted Slope Level Change Estimated Slope Trend Change Time 2005

22 What is Interrupted Time Series Design? intervention intervention before before after after intervention intervention before after before after

23 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Immediacy Time 2005

24 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Lag Effect Time 2005

25 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Permanence Time 2005

26 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Non permanence Time 2005

27 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Time Intervention Introduced Period

28 What is Interrupted Time Series Design? Policy (Intervention) Outcome Pre-Policy Post-Policy Pre-intervention Effect Time

29 AutoRegression Model (One change point) Structural Relation: Error Relation: e t = ϕ 1 e t-1 + ϕ 2 e t ϕ p e t-p +u t Error Assumption: e t ~NID(0, σ 2 )

30 AutoRegression Model (One change point) β0 estimates the baseline level of the outcome, mean number of prescriptions per patient per month, at time zero; β1 estimates the change in the mean number of prescriptions per patient that occurs with each month before the intervention (i.e. the baseline trend); β2 estimates the level change in the mean monthly number of prescriptions per patient immediately after the intervention, that is, from the end of the preceding segment; and β3 estimates the change in the trend in the mean monthly number of prescriptions per patient after the intervention, compared with the monthly trend before the intervention.

31 AutoRegression Model (more than one change point) Policy (Intervention) Outcome Pre-Policy Post-Policy Time Intervention Introduced Period

32 AutoRegression Model (more than one change point) Structural Relation: Error Relation: e t = ϕ 1 e t-1 + ϕ 2 e t ϕ p e t-p +u t Error Assumption: e t ~NID(0, σ 2 )

33 Advantages of Using Segmented Time Series on Evaluate Pharmaceutical Policy Effect Segments Specific event causes a change in the series, dividing it into distinct parts Estimating the changes in the series (level and trend) allows you to assess the impact of the event Strongest quasi-experimental design With or without control group Baseline level and trend serve as built-in control Protects against most alternative explanations Analytic Principle Regression analysis methods Can use short series (<50 data points)

34 Various Models of Time Series Analysis Autoregression Model Distributed Lag Model Simultaneous Equations Model Pooling Model Trend Forecast Exponential Smoothing Model Time Series Decomposition ARIMA/SARIMA/ARIMAT/SARIMAT Box-Cox Transformation Bayesian Integration

35 Contents 1. Introduction of Pharmaceutical Policy Research 2. Interrupted Time Series Design and Segmented Regression for Pharmaceutical Policy Evaluation 3. Examples

36 Example 1.

37 Objectives: To control increasing pharmaceutical expenditures, Taiwan s National Health Insurance has implemented a series of drug reimbursement price reductions since This study examined changes in use and expenditures of oral antidiabetic medications following the price regulation in November 2006.

38 Methods: We obtained claims data between January 2006 and August 2007 from Taiwan s National Health Insurance Research Database. We categorized oral antidiabetic products as affected by the reimbursement reduction ( targeted ) or not ( non-targeted ), by level of relative price reduction, and by manufacturer type (international vs. local manufacturers). We used an interrupted time series design and segmented regression models to estimate changes in monthly per capita prescribing rate, volume, and insurance reimbursement expenditures following the policy.

39

40

41

42 Example 2.

43 Objectives: The purpose of this study is to evaluate the utilization of four approved antihyperuricemic agents in Taiwan before and after two safety announcements rescinded an indication for allopurinol and added a warning on benzbromarone-induced hepatotoxicity in the year 2005.

44 Methods: An interrupted time series design and segmented regression models were used to examine impacts of the safety announcements on the utilization of allopurinol, benzbromarone, probenecid, or sulfinpyrazone. All outpatient prescriptions of the four antihyperuricemic agents were extracted from a longitudinal cohort dataset with individuals randomly sampled from the National Health Insurance Research Database. We examined utilization patterns of antihyperuricemic agents before and after the policy intervention (i.e., safety announcements and labeling changes of allopurinol and benzbromarone) in the year 2005.

45

46

47 THANK YOU! 徐之昇 Jason Hsu, Ph.D. Assistant Professor Institute of Clinical Pharmacy and Pharmaceutical Sciences National Cheng Kung University, Taiwan 國立成功大學臨床藥學與藥物科技研究所助理教授 30th Annual Meeting of the International Society for Pharmacoepidemiology October 23, 2014