Contents Definition of intervention research Characteristics of intervention research Analyses Exercise Reporting Summary
Intervention research Quantitative measurement of effects of therapy or preventive measures Experimental: investigator determines who is treated, not the treating physician
Example: Sehat® Sehat®: new blood pressure drug 1st experimentNy. Ani with high blood pressure Intervention:6 weeks sehat® Outcome:blood pressure
Weekly measured systolic blood pressure of Ny. Ani using Sehat®
Learn from single observation or ‘experience’ “For certain patients the blood pressure will decrease after using the drug” What reasons can you think of that explain Ny. Ani’s decrease in blood pressure?
Learn from single observation or ‘experience’ Explanations for the observed effect: - Regression to the mean -Natural course / prognosis of disease -External effects -Measurement error -True effect drug
Regression to the mean Quartiles systolic blood pressure 1st measurement 2nd measurement Centripetal movement of data in successive measurements Effect of variability “the Doctor’s friend” Solutions: measure more often control group
Natural course / prognosis of disease Independent of treatment, blood pressure can change over time and this change can differ between people
External effects Only interest: effect of Sehat® Effects of other factors can influence the measurement of the effect of Sehat® in two ways: -“placebo” effects - induced effects
Induced external effects Behavioral changes as effect of treatment of high blood pressure with Sehat®, e.g. eating and drinking pattern, physical activity, etc.
Observer effects Patients, treating physicians, manufacturer can have expectations for the effect of Sehat® These expectations can influence, for example the reporting of patients or the measurements made by treating physicians Result: measurement error
Learn from single observation or ‘experience’ Quantitative measurement of the effect of Sehat® can be distorted (confounded) because of : Natural course / regression to the mean (NC) External effects (EE) Observation errors (Information Bias) (IB)
Weekly measured systolic blood pressure of Ny. Ani using Sehat® Observed effect
Components of the observed effect Observed effect (OE) = Drug effect (R x ) + Natural course (NC) + External effects (EE) + Observation errors (Information Bias) (IB) General: only interest in R x
How do we distinguish R X effects from confounding effects? We ‘control’ the experiment
Comparison: Ny. Ani (green) with Sehat®, Ny. Sri (pink) without Sehat® RX?RX?
Controlled study Index (drug treatment): OE i = R x + NC i + EE i + IB i Reference group (no treatment): OE r = NC r + EE r + IB r
Therapeutic drug effect OE i - OE r = R x + (NC i - NC r ) + (EE i - EE r ) + (IB i - IB r ) So, OE i - OE r = R x If, NC i = NC r and EE i = EE r and IB i = IB r
Validity Essence: validity in experimental research is assessed through comparability of groups The basis of an experiment’s design is the prevention of non-comparibility
Comparability of natural course =comparability populations = comparability prognosis What could we do to assure comparability of the natural course?
Clinical trial: “The effect of coffee to daytime drowsiness”
Now open your envelope! coffeeno coffee male2519 Live with parents2117 First child1619 High school in Jakarta2219
Comparability of natural course Options: -Selection or matching -Measure prognostic variables at baseline and control for these during analysis -Randomisation - paradigm for comparative research since 1948
Aim of randomization To ensure that the groups to be compared have the same average baseline probability of change in blood pressure (prognosis, natural course) To make the index and reference groups comparable for all known and unknown factors that may influence blood pressure
Randomized trial Table 1. Baseline characteristics at randomization Sehat® YES Ny. Ani NO Ny. Sri Age (years)4157 BMI (kg/m 2 )24,632,3 SBP (mmHg)160
Randomized trial: Ny. Ani (green) with Sehat®, Ny. Sri (pink) without Sehat® RX?RX?
Randomized trial Table 1. Baseline characteristics at randomization Sehat® YES N = 5,000 NO N = 5,000 Age (years)52,352,4 BMI (kg/m 2 )25,725,6 SBP (mmHg)160
Randomized trial: 5000 women (green) with Sehat®, 5000 women (pink) without Sehat® RX?RX?
Comparability of external effects (EE i =EE r ) What could we do to assure comparability of external effects?
Comparability of external effects (EE i =EE r ) Options: -Randomize -Placebo or simulated treatment in the reference group -Blinding
Comparability of observations (IB i = IB r ) What could we do to assure comparability for the observations (=no observation errors)?
Comparability observation (IB i = IB r ) Options: - Use protocols, systematics - Placebo - Blinding (single, double, triple)
Comparability observation (IB i = IB r ) Need to blind depends on the interpretability of the studied endpoint: - death - myocardial infarction - angina pectoris - blood glucose - quality of life
Measurement Bias - minimizing differential error Blinding – Who? –Participants? –Investigators? –Outcome assessors? –Analysts? Most important to use "blinded" outcome assessors when outcome is not objective! Papers should report WHO was blinded and HOW it was done Schulz and Grimes. Lancet, 2002
Placebo effect Trial in patients with chronic severe itching Cyproheptadine HCL Trimeprazine tartrate No treatment Treatment vs no treatment for itching
Placebo effect Trial in patients with chronic severe itching Cyproheptadine HCL Trimeprazine tartrate Placebo No treatment Treatment vs no treatment vs placebo for itching Placebo effect - attributable to the expectation that the treatment will have an effect
Randomized triple blind placebo controlled trial Table 1. Baseline characteristics at randomization Sehat® BUGAR® N = 5,000 PLACEBO N = 5,000 Age (years)52,352,4 BMI (kg/m 2 )25,725,6 SBP (mmHg)160
Randomized triple blind placebo controlled trial: 5000 women (green) Sehat BUGAR®, 5000 women (pink) Sehat PLACEBO® RX?RX?
Need of comparability Natural course/prognosis: always External effects: depends on aim research: “explanatory” vs. pragmatic Observation errors: depends on endpoint
Intervention: and then Different aims: explanatory vs pragmatic Analysis Loss-to follow-up Choice of study endpoints Disadvantages Alternatives for experimental research Size of trials Reporting
Aims of intervention Explanatory Interest in a single aspect of high blood pressure treatment, by Sehat® Pragmatic Interest in strategy (procedure with all that belongs to it) of high blood pressure treatment, e.g. combination of drugs with living rules and weight loss, including induced effects
Analysis randomized research Randomisation is a powerful way to solve the problem of differences in the natural course: This principle should not be undone in the analysis!
“Intention to treat” analysis Once a member of the cohort, always a member of the cohort contrary to Analysis of only those patients who really received the treatment (‘per treatment’ or ‘per protocol’ analysis). Problem: loss to follow-up
Loss to follow-up: what is the problem? People stop treatment for a reason –Treatment does work or does not –People are ill or are not Reasons can be related to the occurrence relation: drug and outcome Problem: we do not know why people stop Result: bias?
Outcomes and endpoints Intuitive preference for “hard” clinical measures, but: Growing realization of importance of patient preference in assessment of choices Often unclear choice of endpoints Often unclear validity of chosen endpoints
Exercise intervention ( open your exercise book!)
Question 1 Do children of mothers from high risk-families (patient/domain) have a lower 2-year risk of atopic diseases (outcome) if they are exposed, before and after the pregnancy, to probiotics (intervention/determinant) than to placebo (comparison)?
Question 2 Somewhat more atopy and smoking in placebo families, a somewhat more higher incidence of pets and detectable IgE in Lactobacillus, but as a whole reasonably comparable
Question 3 and 4 2-year risk of atopy in Lactobacillus group 15 / 64 = 23% 2-year risk of atopy in placebo group 31 / 68 = 46%
Question 6 E ln0.51 ± 1.96√[49/15*64 + 37/31*68] 95% CI = 0.31 to 0.85
Question 7 Confidence interval tells something about the precision of the effect estimate
Question 8 Fairly strong protective effect of Lactobacillus treatment against developing atopy Complete explanation of effect by baseline differences or differential loss to follow-up very unlikely
Question 9 Pre- and postnatal use of Lactobacillus in high risk children seems to prevent the development of early atopy
Disadvantages of trials Limits to generalisability Selection of the study population Budget RCT is expensive Takes long RCT is prospective Number of patients Ethical dilemma’s (a.o. equipoise)
Alternative: comparative non- experimental research Cohort studies / Case control studies Not inherently less valid, but much more difficult to design and conduct and therefore much more sensitive to bias In comparative non-experimental research exists a large probability of non-comparibility of especially those three components that are solved so well in an RCT
Confounding by indication The prognosis influences the probability to be assigned a certain intervention E.g.-observational studies on the effectivity of vaccinations - observational study on the effect of antihypertensives
Confounding by indication Confounding by indication: research on the effect of anti-hypertensives among 793 Dutch hypertensive women, who were followed for over 10 years Crude and adjusted rate ratios for fatal cardiovascular diseases in treated women compared with non-treated women
Confounding by indication RR cardio-vascular mortality 95% CI Crude 1.00.6 TO 1.5 Adjusted*0.60.3 TO 0.9 *For age, Quetelet index, hart frequency, smoking habits, serum cholesterol, diabetes, previous myocardial infarction or stroke
Reporting randomized trials Table 1: prognostic factors in index and reference group (see exercise) –Show whether randomization succeeded Table 2: shows intervention effects –Difference in group averages, difference in group proportions –Relative risk (reduction), risk difference, NNT
Study size: needed and available number of patients “Sample size calculation” –Alchemy of the statistics –However: prior estimation of information content (precision) of a study based on size is important –Sample size calculation gives global impression –If study quality is insufficient: optimal size 0 –More patients not necessarily more information
Conclusions To evaluate the effects of therapy a comparison is necessary In trials, very effective methods have been developed to enhance the comparability of the natural course, external effects and information randomisation, blinding and placebo The concepts and principles of a trial are a model for non-experimental research