Prof. dr. Davor Eterović EBM-2011/Klinička biostatistika RCT Prof. dr. Davor Eterović EBM-2011/Klinička biostatistika
RCT T –pokus dokazuje kauzalnost C –kontrola male učinke razlučuje od nule, veće mjeri ... R - …bez omaški zbog randomskog usklađivanja
RCT: vrline i mane Najjači dizajn (najpouzdaniji zaključci) Nezamjenjiv za male, ali važne efekte ali ponekad i Teško provodiv, kompliciran, skup Etički dvojben Dvojbene primjenjivosti na praksu (netipični bolesnici, netipični tretmani, preintenzivno praćenje) Zbog toga: Zahtijeva pilot pokus i detaljan protokol: obrazložena hipoteza, plan izvođenja i analize podataka i Hipoteza valja biti vrlo vjerojatna (etički problem kontrola kod teških ishoda; alternativa- nekontrolirani pokus) Većinom ne otkriva novo, već potvrđuje/precizno evaluira, slijedi nakon opservacijskih istraživanja/nekontroliranih pokusa
Kako generirati slijed pridruživanja Jednostavna randomizacija (generator slučajnih brojeva) Korištenje blokova zbog podjednakih skupina Eksplicitna kontrola kovarijabli: stratifikacija ili minimizacija
Kako ne devalvirati randomizaciju Zatajivanje pridruživanja (allocation concealment)- meta analize: vrlo važno, uvijek moguće Maskiranje ispitanika, medicinskog osoblja, statističara; nekad nije moguće ITT (intention-to-treat) analiza, žrtvuje se eventualni lažno negativan rezultat da se ne naruši randomska usklađenost; za nuspojave ne, već- PP (per protocol) analiza
Faze izvođenja RCT Nakon planiranja (pilot pokusa) i dobivanja dopuštenja 1. Izbor ispitanika 2. Mjerenje karakteristika 3. Randomizacija 4. Intervencija 5. Praćenje (evaluacije) ishoda, mjerenja Slijedi izvješće, po strogim pravilima (CONSORT)
Kako izvijestiti rezultate RCT (1) CONSORT guidelines Dijagram toka Karakteristike ispitne i kontrolne skupine: tablica 1. + komentar uspjeha randomizacije, razlike ne testirati formalno (no p-values in table 1!) Tablica 2: Jednostavni, neposredni rezultati ITT analize glavnih ishoda (x+-95%CI) Ako je suradljivost bila slaba i (ili) varirala između skupina, ili ako je bilo dosta izgubljenih podataka, prikaži i PP rezultate
Kako izvijestiti rezultate RCT (2) 5. Ako randomizacija nije perfektna, prikaži i usklađene rezultate: (a) kontinuirane varijable: ANOVA, multipla regresija (b) kategorije: Mantel-Haenszelov test (jedna kovarijabla) ili logistička regresija (više kovarijabli), Poissonova regresija (za stope), Coxova regresija (preživljenje) 6. Ako su planirane/opravdane, prikaži i analize po podskupinama 7. Prikaži nuspojave i neželjene učinke (bez formalnog testiranja; PP prikaz) 8. Analiziraj i (eventualne) sekundarne ishode
Simple 2-arm trial Patients are randomised to study or control group Study population Study Control (50%) (50%) Can have n:m rather than 1:1 allocation E.g. 2:1 active:control
Why extend simple 2-arm RCT? #1: Compare >1 intervention May be the ‘more’ ethical design Can be cheaper to do 1 trial investigating 2 interventions than two separate trials #2: simple RCTs exclude those patients with strong preferences With a chance of getting 1 of 2 interventions more subjects may be willing to be randomised With data on those unwilling to be randomised the trial may be more generalisable #3: Contamination of treatment effects? So instead of randomising a patient, randomise a family, or a GP surgery, or a hospital – cluster randomisation
RCTs for more than one intervention Multi-arm trials Factorial designs Crossover designs
Multi-arm trial Simplest extension to simple RCT Patients randomised to two or more study groups or control group Study population Intervention 1 Intervention 2 Control (33%) (33%) (33%)
Multi-arm trial (2) Advantages: still simple to design allows head to head comparisons Disadvantages: requires a larger overall sample size to achieve the same level of power Multiple comparisons rarely have power to detect significant differences between the interventions
Factorial design (1) Compares more than one intervention Multiple layers of randomisation Notation: 2x2 - indicates 2 trts each with 2 levels 2x2x2 - indicates 3 trts each with 2 levels Fractional factorial designs Many treatments, patients get a selection
Factorial design (2) - 2x2 example Vitamin D and/or calcium supplementation to prevent re-fracture (RECORD)
Factorial designs (3) Advantages: reduced loss of power compared with multi-arm trial very efficient - ‘two trials for the price of one’ allows possibility of exploring interaction effects Disadvantages: requires no interaction between treatments for full power* more difficult to operationalise * There are however studies with a factorial design which specifically anticipate an interaction
Crossover trials Useful when studying patients with a chronic (long-term) disease Allows patients to receive both treatments sequentially “patient acts as their own control” First period B A Second period A B
Crossover trial - example Renal dialysis - each patient receives dialysis 3 times a week Two types of dialysis solution available - acetate and bicarbonate Thought that bicarb may reduce nausea and other symptoms Crossover trial: each patient does a month on one solution followed by month on the other for each patient, the starting solution is assigned randomly
Crossover trials Advantages: requires fewer patients as each get both treatments background “noise” reduced as comparison is within-patient Disadvantages: must be no “carryover” effect Washout periods > 2 periods? Loss to follow up can only be used for short term outcomes e.g. symptom control requires chronic and stable illness - patients require same level of illness for both treatments
Why extend simple RCT - reason 2 Some RCTs compare very different treatments eg surgery vs. long term medication Patients with strong preferences not willing to be randomised Simple RCTs have to exclude those patients
Patient preference trials If patients have a strong preference for a therapy they get that therapy If no strong preference, patients randomised Primary analysis still based on randomised groups Two studies – a randomised study and an observational study
Patient preference trial - example Two treatments for reflux disease: medical management surgical management Four trial groups: prefer surgery prefer medical randomised to surgery randomised to medical
Patient preference trials Advantages: recruitment maximised motivational factors maximised in the preference groups motivational factors equalised in the randomised groups results potentially more generalisable Disadvantages: harder to analyse and possibly to interpret may be unequal distribution across the four trial groups more complex informed consent
Why extend a simple RCT - reason 3 There is a worry that there will be contamination of treatments across patients eg trial comparing two dietary interventions - what if 2 members of same family randomised to different diets? Potential solution - randomise intact groups (families) rather than individuals
Cluster randomised trial Intact groups (known as clusters) rather than individuals randomised to each intervention Unit of randomisation should minimise risk of contamination eg family, practice, hospital ward
A cluster RCT Randomise Providers Control Experimental
Cluster trials - issues Outcomes within a group of patients, or cluster, may be more similar than those across clusters - they are no longer ‘independent’ A statistical measure of this similarity within clusters is the intra-cluster correlation Because patients not independent, study loses power The larger the intra-cluster correlation the larger the inflation required to the sample size to redress the loss of power
Cluster trials Advantages: minimises contamination between groups may be easier to organise practically Disadvantages: requires larger trial patients within clusters not independent standard analysis techniques not appropriate analysis more complex
Different model for randomisation (1) Standard procedure - get informed consent then randomise Potential problems: patients may withdraw if they do not get the treatment they hoped for patients may comply poorly if they get the control treatment - thinking the experimental treatment is better anyway
Different model for randomisation (2) Alternative approach - Zelen’s design: randomise before obtaining consent only seek consent from those randomised to experimental treatment ‘control’ patients not approached for consent Debate surrounds ethics of this approach - eg MRC do not accept this design as ethical
Zelen’s design Advantages: does not raise hopes of a new treatment which can then be denied by randomisation may avoid downward bias in those allocated to ‘control’ Disadvantages: ethics are debateable