Principles and Methods of Intervention Research

Principles and Methods of Intervention Research NVVO, February 2, 2009 Jan G.P. Tijssen, Ph.D. Academic Medical Center - University of Amsterdam

Introduction Pathophysiologic and pharmacological insight Theory or hypothesis about efficacy of a medical intervention Clinical studies to evaluate clinical usefulness Randomised clinical trial has become the paradigm

Clinical decisions Balancing of the magnitude of the Expected therapeutic benefit (= Intended effects) Adverse effects (= Unintended effects) Costs

Clinical Trial A means for measuring expected therapeutic benefit

Example

BMCR trial Patients Previously untreated metastatic colorectal cancer Investigative Treatment Bevacizumab or placebo Background Treatment Irinotecan, Fluorouracil, Leucovorin (IFL) Outcome Median Survival Time (MST)

Results BMCC trial

Results BMCC trial Treatment #PTS MST Surv-1y Bevacizumab 402 20.3 74.3% Placebo 411 15.6 63.4% Treatment benefit Improvement in median survival 4.7 mo Improvement in 1-year survival 10.9% Hazard Ratio for death 0.66 MST = Median survival time (months)

HIVEX trial

HIVEX trial Patients Adenocarcinoma of the mid-to-distal esophagus or of the gastric cardia involving the distal esophagus Treatment Transthoracic esophagectomy with extended en bloc lymphadenectomy Transhiatal esophagectomy Outcome Median Survival Time (MST)

Results HIVEX trial

Results HIVEX trial Treatment #PTS Rate* Transthoracal 114 68 61% Transhiathal 106 74 71% Treatment benefit Improvement in 5-year survival 10% Improvement in median survival 4 mo *at 5 years

Measurement device Object of measurement Measurement errors Systematic errors Random errors

Object of measurement Treatment effect (Only meaningful in relative terms) The disease entity (condition, indication) Treatments to be compared The disease outcome (clinical course)

Disease entity (1) In / exclusion criteria Disease Recruitment scheme Patients with the disease Patient characteristics Clinical spectrum Patients 'represent' the disease at issue

Outcomes BMCC Primary outcome Overall survival Secondary Outcome Progression-free survival Response rate Duration of response Safety Quality of life

Outcomes HIVEX Primary outcome Overall survival & disease-free survival Secondary Outcome Early morbidity & mortality Quality-adjusted life-years Cost-effectiveness

Ordering of outcome parameters Primary outcomes Secondary outcomes

Primary outcome parameter One single parameter related to primary objective Identified in advance Specified in objective of the trial Related to study size calculation Determines positive/negative trial

Secondary outcome parameters Other parameters relating to efficacy Provide supportive evidence May generate hypotheses

Measurement errors Accurate measurement Lack of systematic error Internal validity Precise measurement Lack of random error Precision is based on the size of the treatment groups

Lack of systematic errors When index treatment is as effective as as reference treatment No differences except for random variation Principles of study design derive from the need to avoid systematic errors.

Influential factors Extraneous factors Attention of the physician Psychology of patient Adjustment life style Decisions about other treatments Prognostic factors Risk profile of the patients Information about outcome Investigative treatment

Methods Masking of treatment Patient & treating physician (primary process) Placebo, double dummy, independent physician Randomization Random assignment of study treatment Blinding Blinding of outcome assessor

Comparison OC 1 OC 0 EF 1 = EF 0 NC 1 = NC 0 IE 1 = IE 0 Masking / Placebo Randomization Blinding of assessor TE

Functionality Masking (of patient & physician to treatment) Comparability of extraneous factors Randomization Comparability of groups (prognosis) Blinding of outcome assessor Comparability of information on outcome

Drug trials in oncology Masking of treatments is important Other treatments given must be the same! Wellbeing of patient Adverse effects may prevent masking. Randomization is key Blinded assessment of the outcome is consequence of masking

Results HIVEX trial Treatment #PTS Rate* Transthoracal 114 68 61% Transhiatal 106 74 71% Treatment benefit Improvement in 5-year survival 10% Improvement in median survival 4 mo *at 5 years

HIVEX trial Comparison of strategies No extraneous factors Everything is part of either procedure Comparability of extraneous factors is not an issue Open trial design Randomization is key Blinded assessment of the outcome is critical

Extraneous factors What aspect of the investigational treatment is to be studied? Focus on chemical substance Masking (placebo) is necessary Focus on procedure as a whole (strategy) Open comparison of strategies

Random assignment Purpose Removes investigator selection bias Comparability of known and unknown risk factors Validity of statistical techniques Concealed treatment allocation No knowledge whatsoever in advance about the treatment assignment of a particular patient.

Implementation of randomization In open trial Telephone /computer allocation mandatory Envelopes or lists are subject to tampering Immediate registration by fax or internet Only with double-blind pre-packed medication Systematic randomization is no randomization

Comparability of information Information bias (or detection bias) must be avoided Biased patient reporting Biased ascertainment of information by physician Biased assessment of information by physician, data manager, statistician BLINDING of outcome assessment

Double-blind versus open In 'classical' double blind trial blinding is consequence of masking (placebo). In open trial 'comparability of information' is a separate issue.

Withdrawals Discontinuation of treatment Cross-over to other treatment Non-fatal event Poor compliance Adverse experiences No outcome information

HIVEX trial 109/114 & 93/106 patients underwent the randomized procedure. 4 patients in each group did not have adenocarcinoma in the resection specimen. All patients were analyzed in randomized group.

Withdrawal Usually related to clinical cause Jeopardizes comparability of natural course Infringes on result of randomization Solutions Intention-to-treat analysis Last observation carried forward (LOCF)

Intention-to-treat analysis Principle Outcomes are available in all patients Outcomes are counted according to original randomization, irrespective of actual treatment Consequence Dilution of treatment effects Similar to ordinary clinical practice Preferred method in mortality trials

Availability of Outcomes Yes No Intention-to-treat analysis possible Ad hoc solutions Worse case imputation Leaving patients out from the analysis Missing outcomes must be avoided!

Measurement device Object of measurement Measurement errors Systematic errors Random errors

Precision Outcomes are subject to random variation Precision = random variation in observed effect estimate Precision is quantified via confidence interval

Results HIVEX trial Treatment #PTS Rate* Transthoracal 114 68 61% Transhiathal 106 74 71% Improvement in 5-year survival 10% 95% Confidence Interval (-3%, 23%) * KM at 5 years

95%-confidence interval Range of values surrounding point estimate Interval estimate of the treatment effect Contains true RR in 95% of the applications 95% confident that true RR is embraced

Conclusion HIVEX

Results BMCC trial Treatment #PTS MST Surv-1y Bevacizumab 402 20.3 74.3% Placebo 411 15.6 63.4% Improvement in median survival 4.7 mo 95% confidence interval? P-value < 0.001 MST = Median survival time (months)

Conclusion BMCC

Width of 95%-CI Depends on size of treatment groups Small studies have wide CIs Carry little information Large studies have narrow CIs Carry much information