Register-based research on safety and effectiveness opportunities and challenges Morten Andersen Department of Drug Design And Pharmacology Declaration of interests Participate(d) in research projects funded by AstraZeneca, H. Lundbeck & Mertz, Novartis, Nycomed, Merck Sharp & Dohme and Pfizer Grants paid to the institutions where I have been employed Personally received fees from Atrium, the Danish Pharmaceutical Industry Association for teaching at pharmacoepidemiology courses Professorship in pharmacovigilance supported by the Novo Nordisk Foundation 1
Disclaimer Member of WHO International Working Group for Drug Statistics Methodology International Society for Pharmacoepidemiology (ISPE) Board of Directors European Network of Centers in Pharmacoepidemiology and Pharmacovigilance (ENCePP) Working Group 2 on Transparency and Independence The Nordic Pharmacoepidemiologic Network (NorPEN) Views and opinions expressed in this presentation are entirely my own and I am not speaking on behalf of any organisation or institution Efficacy vs. Effectiveness Trial vs. real world drug use Trial patients younger, more healthy, use fewer other drugs, more compliant, better monitored Efficacy usually overstates benefits The magnitude of this effect differs according to therapeutic area and form of intervention It is an unrealistic expectation that we will have headto-head randomized trails for every intervention and its combinations in every patient subgroup that exactly mimic routine care 2
Efficacy-effectiveness gap (Eichler HG et al. Nature Rev Drug Discovery 2011, Luce et al. 2010 Milbank Quarterly, Haynes 1999 BMJ) Efficacy-effectiveness gap: Drugs often do not perform as well in clinical practice as in the clinical trials that provide the basis for marketing authorization Efficacy: the extent to which an intervention does more good than harm under ideal circumstances (clinical trial conditions) Can it work? Effectiveness: the extent to which an intervention does more good than harm when provided under the usual circumstances of health care practice. Does it work? (and pharmacoeconomics, cost-effectiveness: Is it worth it?) Study designs for comparative effectiveness Randomised studies Explanatory trial, randomised controlled trial Pragmatic trial, large simple trial Randomisation ensures comparability of treatment groups Non-randomised (observational) studies New user cohort studies Prescribing according to usual practice Epidemiologic analysis required to compare treatment groups Meta-analyses 3
Experimental studies randomisation ensures balance, exchangeability Observational studies conscious selection results in imbalance 4
Observational studies channelling, selection bias, confounding New drug preferentially prescribed to patients with more severe disease or more risk factors for adverse reactions New drug looks BAD compared to the old one Confounding mixing of effects Patient factors become confounders if they are associated with treatment choice and are also independent predictors of the outcome Confounder Treatment Outcome 5
Confounding by indication Hypercholesterolemia Statin AMI Confounding by indication Previous MI Hypertension Diabetes Obesity Exercise Lifestyle Statin AMI 6
Two ways to address confounding Confounder Treatment Outcome Propensity score analysis: Goal To identify patients with the identical likelihood of receiving treatment Some will actually receive treatment others will not 7
Propensity score Quantifies probability that a person is treated given his/her observed covariates Probability(Treatment Covariates) Mimics the prescriber s decision process for treatment Estimated from data Multivariable logistic regression Predicted value is each patient s propensity score Given same PS treated and untreated Tend to have same distribution of covariates Are exchangeable Unconfounded comparisons of risk Use of propensity scores in the epidemiological analysis Stratification (quintiles, deciles) Individual matching Weighting Inverse probability of treatment weights (IPTW) Other weights (e.g. SMR ) Control for PS as continuous covariate 8
08/04/2018 17 Challenges in observational comparative effectiveness research Dealing with counfounding, especially confounding by indication and confounding by severity Finding suitable and accessible data sources Dealing with time-varying exposures and timedependent biases Use of proxies high dimensional propensity scores Comorbidity Age Treatment Outcome Proxies may adjust (partly) for unknown confounders The more proxies, the better, high-dimensional PS (Schneeweiss et al. Epidemiology 2009) 9
Clinical trial: Well defined closed cohort Drug A Drug B 0 1 2 3 Time, years Real world: Open (dynamic) cohorts with time-varying exposure Use of Drug A Use of Drug B Non-use 0 1 2 3 Time, years 10
Bias over time with use of preventive treatment compared with non-use RR vs. no drug 1.0 Healthy initiator Healthy continuer Time on drug Healthy user Intention to treat (ITT, baseline exposure) versus time-dependent exposure (TDE) ITT mimics the RCT situation Initial confounding can be adjusted for Less prone to selection bias after initiation Long-term exposure: bias towards null TDE relates to the hazard function Short-term or long-term effects Cumulative effects Latency or induction periods 11
Table 1 Propensity score matching lead to balance of confounders The PS deception: Table 1 for the PS matched cohorts Looks just like Table 1 from a randomised study The confounding factors are nicely balanced between treated/untreated Question: What is the main difference between this observational PS-matched study and a randomised study? Answer: The randomised study is also expected to be balanced between treated/untreated with regard to unmeasured and even unknown confounders 12
Problem in register-based studies: Missing confounder data Lifestyle factors: Smoking, alcohol, diet, exercise, BMI Clinical data: BP, glucose, HbA1c, cholesterol, spirometry etc. Incomplete adjustment: Biased results Solution: data from electronic medical records, patient registries (diagnosis-specific cohorts, quality registries) Mimic the clinical trial to ensure balanced comparisons: Propensity score trimming (Stürmer et al. JIM 2014) 13
Post-authorisation effectiveness studies: Factors that change the effect of treatment Age Co-morbidities Use of concomitant drugs Disease severity and duration Genetic factors Treatment history Drug misuse Decrease benefit Increase harms Factors related to a country or health care system An example: NOAC vs. warfarin in patients with risk factors 14
An example: NOAC vs. warfarin in patients with risk factors An example: NOAC vs. warfarin in patients with risk factors 15
Research questions relevant in the Post- Authorisation (Phase IV) or Real World setting Risk Management Plan Indications Contraindications Comorbidity Recommandations Marketing Information Related interventions Tradition, culture Healthcare System Exposure Prescribing Dispensing Use Outcomes Beneficial effects Adverse reactions Studies of drug utilisation Studies of safety and effectiveness Confounding by... I ve been taking statins for 20 years and was just in hospital with a heart attack. My wife takes no statins and she has no heart problems. Why? 16
Confounding by indication/severity The reason for prescribing the drug is a powerful determinant for the studied outcome Effectiveness more difficult to study than safety in observational designs Some considerations and possible future developments The randomised design is still the gold standard The post-authorisation effectiveness studies (PAES) can give other important evidence on drug effects PAES should not only mimic randomised trials Factors that influence risks (and benefits) should be explored systematically Pragmatic trials and hybrid study designs should be used more extensively (which I have not talked about) Other types of research provide important information about how drugs are used in real life 17
Acknowledgements Parts of this presentation inspired by presentations given at ICPE conferences Til Stürmer John Seeger Sebastian Schneeweiss Articles on specific topics are referenced in the presentation Structured big data in healthcare BIRTH REGISTER MIGRATION RESIDENCE SOCIO- ECONOMY DEATH REGISTER GENERAL PRACTICE PRESCRIPTION REGISTER HOSPITAL REGISTER HOSPITAL RECORDS BIOBANKS CLINICAL TRIALS POPULATION-BASED RESEARCH COHORTS QUALITY REGISTERS 18