Presenter Disclosure Information Soko Setoguchi, MD DrPH Prescription Drug Data: Advantages, Availability, and Access FINANCIAL DISCLOSURE: Grants/Research Support: NIH, AHRQ UNLABELED/UNAPPROVED USES DISCLOSURE: None
Prescription Drug Data: Advantages, Availability, and Access Soko Setoguchi, MD, DrPH Assistant Professor in Medicine and Epidemiology, Harvard Medical School and Harvard School of Public Health Director in Safety and Outcome Research in Cardiology Division of Pharmacoepidemiology, Brigham and Women s Hospital
Pharmacoepidemiology Study of the use and effects of drugs in large numbers of persons. (Strom) Adverse drug events Drug utilization patterns and adherence Drug efficacy and effectiveness Post-marketing surveillance research The goal is to provide information to support the optimal use of medications and better-informed drug therapy decisions.
Outline What types of health services/outcome research questions need drug use data? Different ways of assessing drug use Characteristic of different drug use data Dispensing data from large administrative databases Types of information Advantage Access Making sense of dispensing data in large databases
% Use of the drug within 90 days after discharge from myocardial infarction hospitalization 100 90 80 70 60 50 40 30 20 10 0 Statin Beta blockers ACEI or ARB 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Year Setoguchi AJC 2007
60% 40% 20% 0% Non-black Black P <0.001 for Racial Difference 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Setoguchi AJC 2007 Percent of patients who filled the drug within 90 days after discharge from myocardial infarction hospitalization
1 Adherence to ACE or ARB in Patients with Heart Failure 0.9 0.8 Proportion ACE ACE Lower CL ACE Upper CL Proportion ARB ARB Lower CL ARB Upper CL 0.7 0.6 0.5 0.4 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59
Comparative Effectiveness Research (CER) The conduct and synthesis of research comparing the benefits and harms of different interventions and strategies to prevent, diagnose, treat, and monitor health conditions in real world settings. - Federal Coordinating Council on CER http://www.hhs.gov/recovery/programs/cer/draftdefinition.html
TNFA Increases the Risk of HF Hospitalization Compared to MTX Fully adjusted (all users) Fully adjusted (new users) 4.0 3.0 2.0 1.8 3.1 2.1 3.5 1.7 2.3 1.0 0.0 Previous HF + Previous HF - Two groups Setoguchi et al AHJ 2009 combined
Types of Health Services/Outcome Research Questions Using Drug Data Descriptive or analytic studies assessing use, initiation, and adherence/compliance of/to prescription drug use Comparative/clinical effectiveness research involving medical treatments The Federal Coordinating Counsil definition include effectiveness and safety Drug use as an exposure
Characteristics of Drug Use Longitudinal and time-varying Patients are not great source of information inaccurate (do not remember, remember incorrectly, or may not tell you the truth) Inaccuracy (misclassification) may be differential Difficult to predict Multiple players (physicians, pharmacists, patients, family etc) Little variability explained by usual patients and provider characteristics
How Prescription Drugs Are Taken by Patients? What has to happen for patients to be able to take a prescription drug?
Clinicians write prescriptions and record in charts Patients go to pharmacies and pharmacists dispense prescription drugs Patients take medications
Different Ways of Assessing Drug Use Obtain Information on Prescriptions Medical records Electronic record of prescribing Obtain Information on Dispensing Drug files in most large databases Record # of times that patients open the bottle Patients or someone count pills Measure blood level of medications Ask Patients
Existing Databases Contain Different Types of Drug Use Data Prescriptions Electronic medical records, electronic prescribing data, clinical registries, GPRD Dispensing Administrative databases (Medicaid, Medicare Part D, State Pharmacy Assistance Programs, large HMOs) Bottle openings Not available in databases Pill Count Blood level Ask Patients Not available in databases Not available in databases NHANES (and see bottles or meds), population based cohort studies (Framingham, Nurses Health study etc)
Pros and Cons of Different Drug Data Methods Pros Cons Prescriptions Dispensing Bottle opening, blood level etc Ask patients Good source if you want to know what is prescribed (physicians behavior) vs. what is used by patients Available in large administrative databases, longitudinal, closer to actual intake than prescription data Highly accurate Can obtain additional information on reasons of use and low compliance Information is sometimes missing in charts (West SL J Clin Epi 1994) Further from actual intake No information on indications No information on actual intake Very expensive to collect data Inaccurate, differential recall by exposure or outcome status
Self-Reported Drug Use Patients Enrolled in a HMO for >12 years and Aged 50-80 years. 57% patients recalled using NSAIDs (41% of one time user and 85% of repeated users) 30% recalled name and 15% recalled both name and dose for NSAIDs 78% recalled name and 28% recalled both name and does for estrogen West et al. Am J Epi 1995
Dispensing Data from Large Administrative Databases
Advantages of Prescription Drug Data in Administrative Databases Considered gold standard Gold standard does not indicate perfect or 100% accurate. Large Longitudinal Detailed can calculate daily dose, duration of use, adherence with some assumptions
Information Typically Included Drug Files in Administrative Databases Date of dispensing National drug codes Generic name Form Strength Manufacture etc. Quantity dispensed Days supply Prescriber ID Cost
National Drug Codes (NDC) Universal product identifies for human drugs 10 digits number Updated monthly Limited to prescription drugs and insulin products May not be on the list the firm has notified the FDA that the product is no longer being marketed the firm has not complied fully with its listing obligations and therefore its product is not included until complete information is provided.
NDC Can Tell you Dosage form Route of administration Active ingredients Manufacture Strength Package size and type Major drug class
Information Not Included in Drug Files in Most Administrative Databases Indication What is actually prescribed Overall error rate of not dispensing what is prescribed is 0.7% in Ontario Drug Benefit Databases (Levy et al Can J Clin Pharmacol 2003) Inpatient drug use Medical charts Premier Perspective Databases Inpatient administrative databases covering ~1/6 of US hospitalizations
Prescription Drug Data Example
Access to Drug Data from Large Databases with Drug Files Medicaid Medicare Part D State Pharmacy Assistance Programs United Health or BCBS National Health Insurance Data (Canadian provinces) Databases National Health Insurance Research Database (Taiwan) HIRA (Korea) GPRD or THIN (UK) How to Access the Data Purchase from CMS Purchase from CMS Negotiate/Collaborate with each program Purchase available Only through collaboration Only through collaboration Only by Korean researchers Purchase available
Making Sense of Drug Data
Longitudinal Dispensing Data NDC NDC description Disp_dt days_ supply qty_ dispensed 62794014501 Digitek 125mcg Tablet 2-Feb-04 30 30 62794014501 Digitek 125mcg Tablet 12-Mar-04 30 30 62794014501 Digitek 125mcg Tablet 7-Apr-04 30 30 62794014510 Digitek 125mcg Tablet 27-Jul-04 25 25
Creating A Diary
Creating A Diary
Creating A Diary
Prevalent Users Underascertainment of events that occur early in therapy Depletion of susceptibles survivor bias Prevalent users are adherent patients Adherence bias
Survivor Bias Prevalent users are those who survived through events that could have happened soon after the initiation of the therapy Prevalent users may be very different from new users
New-User Design All patients in a defined population (in terms of who and from which period) who initiated a treatment with the study medication Follow-up for endpoints begins at the time or initiation The study patients are further restricted to those with a minimum period of nonuse (washout) prior to the start of the drug Consider pharmacology of the therapy 6 or 12 months has been used
Health User Bias Patients with Preventive medications Newer more expensive medications Less frail Seek more care Follow healthcare providers recommendations Live healthier life style Have more desire to live
Compared to what?- importance of comparison groups Statins are wonder drugs in nonrandomized studies 50% reduction in hip fractures Improvement in cognitive status Reduced cancer risk How choice of comparison group affect the results Statins and cancer outcome
Glynn Epidemiology 2001
Statin and Cancer Linked Medicare-State pharmacy assistance-cancer registry New Users Design Chronic users Active comparison group with similar patients (statin users vs. glaucoma drug users) Exposure Definition: 3 prescription for statin during the first 180 days after the initation of statins Comparison: Initiators of glaucoma drugs with 3 prescription for glaucoma drugs during the first 180 days after the initation Setoguchi, Circ 2007
New Users Design, Chronic Users, Active Comparisons Comparison Glaucoma drug users Adjustment Colorectal Cancer Lower HR 95% CI Upper 95% CI Unadjusted 0.97 0.70 1.34 Sex, agesq, race adjusted** 0.96 0.69 1.33 Multivarible*** 0.97 0.70 1.35 Antihypretensive drug users Non-statin lipid lowering drug users Unadjusted 1.06 0.76 1.48 Sex, agesq, race adjusted** 1.31 0.93 1.84 Multivarible*** 1.34 0.95 1.89 Unadjusted 0.82 0.42 1.61 Sex, agesq, race adjusted** 0.83 0.43 1.64 Multivarible*** 0.83 0.42 1.64 Setoguchi, Circ 2007
Confounding by Severity Are ACEIs effective preventing MI in patients with hypertension? Everybody has hypertension ACEIs vs. Hypertension with Diabetes and proteinuria vs. No ACEIs Uncomplicated Hypertension
Comparative Effectiveness Study Comparing two drugs with similar properties Everybody has hypertension Everybody has similar severity ACEIs vs. Hypertension with Diabetes and proteinuria vs. ARBs Hypertension with Diabetes and proteinuria
Controlling Confounding in Observational Studies Confounding Measured Confounders Unmeasured Confounders Design Restriction Matching Analysis Standardization Stratification Multivariate regression Propensity scores Marginal Structural Models Unmeasured, but measurable in substudy 2-stage sampl. Ext. adjustment Imputation Propensity score calibration Unmeasurable Design Cross-over Choice of comparison (active comparison) Analysis Instrumental variable Schneeweiss PDS 2006, modified by Setoguchi
Summary Dispensing data are often considered gold standard (the best available) data for drug use information Data are detailed, longitudinal, and time-varying Assumptions are needed to create a diary of drug use based on the data Careful considerations in design and analytic methods are needed for a valid study