Challenges in Evaluating Screening & Prevention Interventions Jack Cuzick Cancer Research UK
Screening & Prevention Trials Large Multicentre, International Long Compliance, contamination Expensive Consent Trial Mechanics
Classical Approach Randomised Population Based Trial Intent-to-Treat Analysis of Mortality Strengths Unbiased Directly applicable to population Weaknesses Expensive follow-up Requires high compliance, low contamination
Simplifications Make Trials More ‘Routine’ Surrogate Endpoints Target High-Risk Individuals
Informed Consent Can require more time than intervention Should be relaxed when comparing new intervention at least as effective as conventional Need for ‘Community Consent’ Data Protection Act Research vs Implementation Studies
Cluster Randomisation Can minimize consent issues Requires well-defined population Good compliance essential Preselect based on likely compliance if possible Aim for >60 ‘matched’ cluster Minimise between cluster variation
Reciprocal Trials Avoids issues of untreated controls Two outcomes for same exposure Ex-smokers CT screening for lung cancers vs Aggressive cardiovascular interventions (BP and cholesterol) Two unlinked types of screening Colorectal vs (ovary/prostate)
Compliance (& Contamination) Acceptance of other screening on offer Pre-questionnaires Run-in procedures Contamination Availability of intervention Positive correlation with compliance
Selection of Compliant Population Before Randomisation Strengths Increase of power Efficiency of trial resources Weaknesses Generalizability Greater risk of contamination
Analysis allowing for non-compliance Strengths Estimate of screening effect in compliers Extrapolation to different levels of compliance Appropriate confidence intervals (larger) Weakness Auxillary to ITT population analysis
Randomised Option Contaminators Treatment Control Insist (Group A) Contaminators Accept only if offered (Group B) True Comparison Group Patient’s behaviour regarding treatment Refuse (Group C) Non - compliers
A Hypothetical Example – Binary Outcome
Potentially Dangerous Modifications Non-randomised comparisons Historical Controls Compliers vs Non-compliers Case-Control Studies Survival or stage changes in screen detected cancers Lead time bias Length bias Subgroup Analysis
Phased Introduction Group 1 Group 2 Group 3 Group 4 Group 5 x x x x x Ignore subsequent cancers Group 1 Group 2 Group 3 Group 4 Group 5 x x x x x x x x Phasing period > lead time Prevaluated screen problematic if incidence age dependent must include prevaluated screen Ideally exit screen for all (e.g. groups 1, 3, 5)
Evaluating Service Screening Case-Control Audit Focus on screening failures Cervix – stage Ib+ Breast – deaths (or stage 2) Colon – Duke’s B or greater Compare screening histories of failures (cases) to programme in general (controls) Require well-defined target population Evaluate Coverage Screening interval Follow-up Misreading True ‘false negatives’ Problems with screen-detected cancers
Evaluating Service Screening Modelling – Process Parameters Need surrogate for mortality reduction Previous trials for validation Early (easy) vs late (hard) markers High detection rate of early lesions Reduced detection of advanced lesions on subsequent screens Reduced interval cancer rate Reduced overall rate of advanced lesions
Intermediate Endpoints/Biomarkers Strengths More power Earlier results Modelling of different strategies Weaknesses Need for validation Treatments specifically armed at biomarker
Surrogates for Breast Cancer Risk Mammographic Density Oestradiol Insulin-like growth factor II (?) Cytology Weight (loss)
Risk-Benefit Ratio Critical in Screening or Prevention – well population Side-effects early and patient-specific Benefits late and non-individual specific Costs Individual - Travel, time off work, anxiety (reassurance) Health System - Screening Test Further Evaluation Treatment - (Potential Cost & Reduction) Programme Management & Evaluation