Use of IMS Data in Pharmaceutical Policy Analysis Andreas Gieshoff Vice President Statistics & Advanced Analytics, EMEA Utrecht, January 2010 IMS Health 2008
AGENDA Data Collection Methodologies Sampling and Projection Quality Control Selected Countries
Data Collection Methodologies -Constituting Elements-
Data Collection Methodologies -Study Objectives- Channel Variations Retail Sales to Pharmacies Pharmacy Sales to Consumers Hospital Sales to Hospitals Product usage within hospitals Measurement of Demand Measurement of Supply
Data Collection Methodologies -Study Objectives- Doctor-Patient Interactions Variations Doctor related Medical Data Index (MDI) Diagnosis-Treatment Relations Prescribing speciality Patient related Anonymized Patient Level Data (APLD) Prevalence Disease Pathways Compliance Longitudinal Prescription Data (LRx) Therapy switch Compliance
Data Collection Methodologies - Information typically captured (Sales Data)- Wholesalers/ Distributors Pharmacies Hospitals Product Price / Value Purchasing account / segment Transaction quantity Transaction type o Sale o Bonus o Return Product Price / Value Sell-in / Sell-out quantity Sell-out type orx ocash Product Price/Value Sell-in / Consumption quantity Speciality ward
Data Collection Methodolgies - Information typically captured (Medical Data) - Diagnosis ICD10 codes Doctor wording Co-diagnoses Treated/untreated Patient Demographics Age Sex Smoker/Non-smoker Insurance Doctor Demographics Age, sex Speciality Year qualified University Region Therapy Product prescribed Desired effect Co-prescription ATC, NDF Dosage data
Data Availability Microsoft Office Excel-Arbeitsblatt Microsoft Office Excel Worksheet Sales Data Retail supply data available in all IMS countries (exception: China) Retail demand data available in EU5 and some smaller EU countries Hospital data is mostly supply data IMS covers most relevant channels, but not all Medical Data 43 countries covered (18 Europe, 9 Americas, 8 APAC, 8 MENA). The specialty coverage differs by country. Cross-country analyses should consider this. Data covers physicians in private practice (exception: Belgium has also a hospital MDI).
Price Levels in selected Countries Country Price Level Remarks Brazil Pharmacy Purchase Price List Price China Hospital Purchase Price Weighted Average Price India Stockist Purchase Price List Price Indonesia Pharmacy Purchase Price List Price Pakistan Pharmacy Purchase Price List Price Peru Pharmacy Purchase Price Weighted Average Price Russia Pharmacy Purchase Price Weighted Average Price Egypt Public Price List Price Jordan Pharmacy Purchase Price List Price Morocco Pharmacy Purchase Price List Price Vietnam Ex-Manufacturer Price Weighted Average Price Price levels differ across countries Direct comparisons may be inappropriate
Sampling and Projection -Key elements of sampling concepts- Study Objective Practicality Sample Design Cost Accuracy The right balance determines the relevance of our measurements
Sampling and Projection -Sample Design Stratification- Retail Hospital Doctor-Patient Data Region Size (e.g. turnover) Type (e.g. chains, cooperatives) Region Ownership Specialization Size Region Doctor Speciality Practice Type Accurate universe information is key for precise sampling
Sampling and Projection -Projection Methodologies- Projection factors are typically the quotient of universe elements (N) by sample elements (n) N PF= a n There are variations of complexities Weighting variables (e.g. turnover, size of hospitals) Ratio estimators Geo-spatial projections Regression-based universe estimations The right choice is based on available universe information
Data Collection Methodologies -Data Source Configurations- Data Source Data Source Configuration Market Supply Retail Hospital Pharmacies Wholesalers Distributors Single source Multi-source Market demand Retail Hospital Pharmacies Hospitals Generally single-source Doctor-Patient Interactions Doctors for MDI Doctors for APLD Pharmacies for LRx Generally single-source
Data Quality Error Components Random Error: Sample size Stratification Selection Systematic Error: Non-response Incomplete reporting Reporting time Reporting quality Quality controls minimize the systematic error We can calculate the random error by statistical formulas
Quality Controls Input Process Output Data Formats File Sizes Duplications Completeness Statistical Outliers Confidence intervals normal, gamma, poisson) Expected volumes Box-plot Multivariate controls Cluster analysis Chi-sqare / Mahalanobis Rank&volume correlations Market trend Annual Validations (Bias, Precision)
Bias (only for sales data) Average over/underestimation of the real market performance: Total IMS units of all validated product forms Total real units of all validated product forms Example: Pack IMS Units Real Units R-Value A 1,000 900 1.111 B 1,200 1,500 0.800 C 4,000 D 6,500 3,800 7,000 1.053 0.929 Bias = -3.4% E 7,200 7,400 0.973 Total 19,900 20,600 0.966
Precision Index (only for sales data) Example of Precision R-Value Class from 0.475 0.575 0.675 0.775 0.875 0.975 1.075 1.175 1.275 1.375 1.475 Total to 0.575 0.675 0.775 0.875 0.975 1.075 Precision Index = 1.175 1.275 1.375 1.475 1.525 No. of R-Values 15 35 55 230 590 770 410 100 45 25 5 2,280 800 700 600 500 400 300 200 100 0 0.475 R-Value Distribution 0.575 0.675 Σ = 2,070 0.775 0.875 0.975 2,070 2,280 1.075 1.175 1.275 2,070 Precision Index = * 100 = 2,280 1.375 1.475 R-Values inside ±22.5% deviation range R-Values in total 90.8%
Quality Controls for Medical Data Sample: Design fulfillment Weekly distribution diagram Reporting time Plausibility checks (volume) Data: Completeness control Coding control Plausibility checks (contents) Univariate and multivariate outlier detection
Sampling Error Performance MDI -Morocco (n=5000 patient records, cluster effect=30%)- 16,0% 14,0% 12,0% Sampling Error 10,0% 8,0% 6,0% 4,0% 2,0% 0,0% 45,0% 40,0% 35,0% 30,0% 25,0% 20,0% 15,0% 10,0% 5,0% Measured Share (e.g. of a disease) The sampling error increases with increasing granularity of analysis The distribution of events across sample doctors can alter the accuracy
Estonia - Channels of Distribution - Status 2008 Manufacturers Wholesalers 88% 12% Retail Pharmacies Hospitals
Estonia - Retail Data - Data Source - Wholesalers/Pharmacies - Timeliness Performance -Average DAP - Universe (2009) 528 WS-Coverage 48 % Pharmacies (Rx/LRx) n=227 Type 2008 2009 MIDAS 24 22 On-Site Database System 22 20 Printed 23 23 Frequency Data Quality Measurement Printed On-Site Database System Quarterly Monthly Year Bias Precision 2007 3.1 83.0 2008-8.6 84.6 Updated: Dec-09/mb
Malaysia - Channel of Distribution - Status 2009 Manufacturers / Agents / Distributors 21% 26% 53% Dispensing Doctors Drugstores Retail Pharmacies Others Hospitals Government Private 21% 1% 20% 5% 37% 16% Consumers Market covered by MPA: 69%
Malaysia - Retail Data - Data Source - Retail Sample - Timeliness Performance -Average DAP - Universe (2006) 9,030 Sample (2006) 211 Stratification Reg, Spec, Hosp WS-Coverage 65 % Design Fulfillment 90 % Frequency Type 2008 2009 MIDAS 51 48 On-Site Database System 47 41 Printed 51 46 Data Quality Measurement Printed Quarterly Year Bias Precision On-Site Database System Monthly Report Information 2006-16.7 79.1 2007-13.8 79.5 Report Type Reporting Level Purchase National Updated: Dec-09/mb
IMS Data for Pharmaceutical Policy Analysis IMS data offers a multitude of opportunities for pharmaceutical policy analysis Collected sales data provide facts for pricing and reimbursement analysis IMS medical data measures changes in prescribing habits APLD data allow cohorting of data to measure health outcomes LRx data provide insights into therapy switches Country-specific differences in IMS data are relevant IMS covers most relevant channels but not all Measured prices differ across countries Quality of data differs
IMS Data for Pharmaceutical Policy Analysis - Selected studies conducted by EMEA stats services - Effects of pharmaceutical innovation and demographic change on the German oncology market Study showed effects of product age, innovation, health care reforms, and parallel imports on oncology prescriptions Combination of IMS data (IMS NPA, IMS LifeCycle) and public domain demographic data (GENESIS) Study used random and fixed effects models Factors influencing reimbursement and prescription decisions on the RA-market in EU5 Study showed effects of price and RA prevalence on RA-related prescriptions The study used IMS MIDAS data GLM, logistic regression, and mixed model approaches were used