the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Studi clinici longitudinali: metodi di analisi statistica Massimo Borelli 21 aprile 2016 Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach 2005, John Ioannidis Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach It is simply no longer possible to believe much of the clinical research that is published, or to rely on the judgment of trusted physicians or authoritative medical guidelines. I take no pleasure in this conclusion, which I reached slowly and reluctantly over my two decades as an editor of The New England Journal of Medicine. 2009, Marcia Angell Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach 2015, Richard Horton The case against science is straightforward: much of the scientific literature, perhaps half, may simply be untrue. Afflicted by studies with small sample sizes, tiny effects, invalid exploratory analyses, and flagrant conflicts of interest, together with an obsession for pursuing fashionable trends of dubious importance, science has taken a turn towards darkness. Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach 2016, Ron Wasserstein Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach what does it mean ’longitudinal’ ? 0 2 4 6 8 1012 PaO2 FiO2 hfpv 400 300 200 100 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● control ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 2 4 6 8 1012 time (hour) U. Lucangelo et al., 2011. Early Short-Term Application of High-Frequency Percussive Ventilation. Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach clinical topics Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach the vitrectomia dataset 979 980 981 982 983 984 Soggetto 164 164 164 164 164 164 979 980 981 982 983 984 PTrattato 12 8 12 15 13 11 Nascita 11760 11760 11760 11760 11760 11760 Datavisita 41908 41948 42008 42098 42283 42352 PControllo 12 12 12 12 12 13 Massimo Borelli Sesso M M M M M M Tipovisita Apreop B30 C90 D180 E365 Finale Occhio OS OS OS OS OS OS Intervento Mer Mer Mer Mer Mer Mer Gauge 27 27 27 27 27 27 Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach the maculopatia dataset Subject 1 2 3 4 5 6 .. 280 281 282 AcVis 0.63 0.40 0.20 0.20 0.25 0.63 .. NA NA NA Time 0.00 0.00 0.00 0.00 0.00 0.00 .. 72.00 72.00 72.00 Gender F F F M F M .. F F F Massimo Borelli Age 68 82 71 64 83 79 .. 80 85 88 Eye sx dx dx dx dx dx .. sx sx sx Drug Beva poli Ranib poli poli poli .. Ranib Beva poli Injection 15 14 18 25 7 8 4 3 3 9 Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach time profiles Differenza di Pressione 5 0 -5 0 500 1000 1500 2000 2500 Tempo post-operatorio (giorni) 1.0 Acuità Visiva 0.8 0.6 0.4 0.2 0.0 0 6 12 18 24 30 36 42 48 54 60 66 72 Massimo BorelliFollow-up (mesi) Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Table of contents 1 the wrong analysis 2 the difficult question 3 the mixed-effects models 4 insight: the bayesian approach Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach a wrong analysis Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach a wrong analysis Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach a wrong conclusion The second order polynomial regression exhibit a stronger R 2 determination coefficient (0.30 vs. 0.02 in the linear case), therefore we deduce that the terapy has an effect in slowing down the disease wrong! Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Why that analysis is wrong? we have to face different difficulties to ’reduce repeated information’ into one number assure reliable inference managing the twins effect :-) managing the latent variables / hierarchical structure Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach the twins effect :-) Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach @p − value p − value = 0.54 Massimo Borelli p − value = 0.02 Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach explanation of the phenomenon m1 − m2 t=q 2 s1 s22 + n1 n2 Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach the most dangerous equation Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach the most dangerous equation Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach a wrong conclusion so, what was wrong? to have considered each point as an ’independent’ observation, forgetting that there is information carried in (i.e. previous patient conditions) Massimo Borelli The second order polynomial regression exhibit a stronger R 2 determination coefficient (0.30 vs. 0.02 in the linear case), therefore we deduce that the terapy has an effect in slowing down the disease Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach 8 10 a wrong conclusion Massimo Borelli 2 0 to have considered each point as an ’independent’ observation, forgetting that there is information carried in (i.e. previous patient conditions) 4 6 so, what was wrong? 0 2 4 6 Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach the mixed-effects models good news a mixed-effects model allows to obtain (population) time-evolution estimates from the (random sample) observations fixed effects to take in account the patient-level time-evolution within the (random) sample observed random effects Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach what are we going to talk now? 1 the idea behind a mixed-effects model 2 to explain the difference between fixed and random effects 3 hard – to pursuit a proper model selection Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach 0 0 2 2 4 4 6 6 8 8 10 10 the idea behind 0 2 4 6 y = mx + q + ε Massimo Borelli 0 2 4 6 y = (m + β)x + (q + α) + ε Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach fixed effects vs. random effects y = mx + q + ε y = (m + β)x + (q + α) + ε m, q are (population) fixed effects α, β are (patient) random effects ε is the residual random effects α, β, ε ∼ N(0, ...) cor (α, β) = ... Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach hard - model selection / fixed effects LINE costant 0 2 4 6 10 8 6 4 2 0 0 0 2 2 4 4 6 6 8 8 10 10 parabola 0 2 6 4 6 4 6 5 5 0 -5 -5 2 2 constant 0 0 -5 0 0 line 5 PARABOLA 4 0 2 Massimo Borelli 4 6 0 2 4 6 Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach 10 8 6 4 2 0 0 0 2 2 4 4 6 6 8 8 10 10 hard - model selection / random effects 0 2 4 6 0 (m + β)x + (q + α) cor (α, β) = ρ ε 2 4 6 (m + β)x + (q + α) cor (α, β) = 0 ε Massimo Borelli 0 2 4 6 mx + (q + α) ε Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach hard - model selection Three way are commonly exploited to pursuit a model selection - deviance analysis (under Maximum Likelihood estimates) - information criteria (e.g. AIC) + parametric bootstrap Note: first two methods properly work only on fixed effects J. Faraway, 2016, ISBN 9781498720960. Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach hard - parametric bootstrap Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach hard - model selection Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach with R: model summary Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach with R: to understand summary y = mx + (q + α) + ε Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach insight: the bayesian approach Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach please, note the differences (1/3) wrong1 = lm(AcVis ∼ 1 + Time) summary(wrong1) Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach please, note the differences (2/3) mixed1 = lmer(AcVis ∼ 1 + Time + (1|Subject)) summary(mixed1) Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach please, note the differences (3/3) formula = AcVis ∼ 1 + Time + f(Subject, model = iid) output = inla(formula, family = gaussian) summary(output) Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach great advantage 5 10 0 ~ p(β1|y) posterior marginal -0.7 -0.6 -0.5 -0.4 -0.3 β1 Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach some textbooks Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica the wrong analysis the difficult question the mixed-effects models insight: the bayesian approach ringraziamenti Arjuna, Federico, Roberto: youtube.com/medicinatrieste Massimo Borelli [email protected] www.dmi.units.it/borelli/ Massimo Borelli Studi clinici longitudinali: metodi di analisi statistica