T hinking statistically - PowerPoint Presentation

Slide1

Thinking statistically

Eric Ohuma

Senior Medical Statistician

University of Oxford

Centre

for Statistics in

Medicine & Centre for Tropical Medicine and Global Health

Slide2

A

lui

che le vigilanze, tutto è rivelata.

To one who watches, everything is revealed.Italian proverb.

S

cience

is built upon observations.

Slide3

question

design

data

analysis

interpretation

Study timeline

PUBLICATION!!!!!!!

….but what if things went wrong?!?

Slide4

Aims of researchClinical and epidemiological research studies may have various aims:

To quantify (e.g. prevalence of a disease in community)

To compare (which intervention is better?)

To predict (who gets cancer?)To assess association To explore aetiology (exposure causing outcome)…

Slide5

Main elements of research Clear/precise question(s)

Research Design

Who to study

What to measure and whenWhat interventions to make, and when (if any)How large a sampleMany difficult decisions, so we need a

protocol

Slide6

Measurement “It is difficult to think of anything in medicine that does not, directly, or indirectly, depend on counting or measurement.” Mainland (1952)

“Scientists ought to view their general research objective as one of measurement” Poole et al. (1984)

“Measurement error is one of the most important considerations in clinical medicine, but the effect of measurement error on clinical practice continues to be underestimated.” McDonough (1997)

6

Slide7

Measurement: Factors that can affect measurements Quantity being measured varies

e.g. blood pressure, ejection fraction

The operator

Between and within operator variability Skill/experience of operator EquipmentWith-machine errorBetween-machine variation Ambient factors e.g. temperaturePatient’s behaviour (e.g. recent diet, exercise)

Digit preferenceKnowledge (e.g. clinical history, previous measurement) 7

Slide8

Exercise

8

Slide9

The example of Zika-virus and

microcephaly

researchEmergency, feasibility, causality

: constraints on the design of epidemiological studies

Slide10

Context

Zika

virus: First

discovered in Uganda in 1948Transmission by mosquito bite (mainly

Aedes spp.); sexual transmission is also reportedDengue-like fever illness, with associated neurological symptomsRecent spread of the virus since 2010: - reported to cause outbreaks in several islands in the Pacific ocean (Micronesia,

Polynesia

…)

- identified in several African and Asian countries - causing a major outbreak in Brazil, and more largely in Latin America and Carribean since 2015An international public health emergency

was declared in 2016, related to the description of neurological defects in newborns (microcephaly) associated with Zika infection during pregnancy.

Slide11

Context

Recent

spread of the virus

since 2010: - reported to cause outbreaks in

several islands in the Pacific ocean (Micronesia, Polynesia…) - identified in several African and Asian countries - causing a major outbreak in Brazil, and more largely in Latin America and Carribean since 2015An international public health emergency was declared in 2016 by WHO related to the description of neurological defects

in

newborns

(microcephaly) associated with Zika infection during pregnancyBefore 2015, almost nothing was known or reported on Zika virus and the

consequences of infection during pregnancy on central nervous system.

Slide12

Zika: What

is

known so far…Zika

virus literature search in Pubmed: - 50 articles between 1948 and 2010 (1 report of CNS infection in mice) - Only 13 cases reported until 2007 - 70 articles between 2010 and 2015 (10 articles with keywords neurologic/nervous/microcephaly/guillain) - 1130 articles since January 2016 (468 articles with keywords neurologic/nervous

/

microcephaly

/guillain)

Slide13

Question

Is

Zika

virus infection of the mother during

pregnancy causing neurological defects in newborn babies?

Slide14

Question

Is

Zika

virus infection of the mother during

pregnancy causing neurological defects in newborn babies? This example will provide an insight of epidemiological research on causality progressing and in an emergency context… what are the processes, what are the types of studies that

answer

to different constraints and how does research progress towards always stronger evidence ?

Slide15

Group ExerciseYou are the country representatives on emerging infections & global health and have

been asked to design a study to help understand

whether there is an association or causal pathway between

Zika virus during pregnancy and neurological defects among newbornsIn the next 45-60 minutes, discuss in your group about how you would go about it and the potential limitations of the chosen design and any specific country challenges you may face in answering this questionPrepare a 15 minute summary to present to the group.

15

Slide16

Group ExerciseGet into your respective groups (4):

Group

Ivory Coast: Case-control

Group Uganda: RCTGroup Ethiopia: Cohort studyGroup Ghana: Case-series

16

Slide17

Group Exercise: Things to Consider

Purpose:

- what do we want to learn (think about the question)2. Target population: - Research is conducted on a sample of individuals

- The sample should be representative of a population3. Selecting the participants - Case definition - Inclusion criteria - Exclusion criteria (reasons for excluding some (few?) participants) - can we define and measure all that needs to be defined and measured?Note that degree of selectivity affects inferences about the population (generalisability)

17

Slide18

Group Exercise: Things to Consider4.

Recruitment

processes

- What information should be collected? - How quickly is it needed? - How will you act on the results?5. Analytical considerations

- sample size considerations i.e. what aspects you need to calculate this6. Limitations of the study - Limitations specific to the question and design7. Assessment of the level of evidence for your study8. Some recommendations/conclusion18

Slide19

Case series

Review

available dataRetrospective

IncompleteClassification issues- Clarify the diversity of manifestationsFrança et al, Lancet 2016

Slide20

Case-control study

Lancet 2016

design

case-control study

prospective8 public hospitals in Recife, BrazilOutcome=microcephalycases: neonates with microcephaly. 2 controls/case: neonates without microcephaly, matched by expected date of delivery and area of residenceExposure=Zika virus infectionExposed: presence of Zika virus-specific IgM or a positive RT-PCR in neonate serum sample.Possible confounders=Mother factors: education, age, exposure to Zika?Matched for date and place=

exposure

to

Zika, other environmental factorsresults32 cases; 62 controlsCases Controls OR (95%CI)

Slide21

Cohort study

design

cohort study

prospective1 hospital in Tegucigalpa, HondurasInclusion=Pregnant

women upon their first antenatal consultationExposure=maternal Zika virus infectionExposed: suspected (fever + other clinical symptoms); probable/confirmed (presence of Zika virus-specific IgM, different tests)Outcome=microcephalydefined as an occipito-frontal circumference <2SD for sex and gestational age, using the INTERGROWTH-21st chartsAnalysis planned so far (because

lack

of means: case-cohort):samples of all mothers of a microcephaly casesamples of 200 mothers of a ‘normal’ baby in order to estimate prevalence/incidence of Zika-v infection in the general population

Slide22

Timeline

of

presented

studies201320142015

2016

French

Polynesia Zika

outbreakSept13-Mar14March 16Math. modelling French Polynesia*Data collection for case-controlJan16-May16

Sept 16Case-controlExcess of microcephaly cases detected in BrazilOct15Brazil Zika

outbreakMar15-Sept15…July 16Surveillance analysisJune 16Ecological correlation,Case-series,Cohort study

protocol

Feb 16Case reportMay 16Case report* The excess of microcephaly cases could only be detected in the French Polynesia outbreak in the light of the situation in Brazil.

May 16Review on causality

Slide23

Bias in observational studies

Selection bias

In a cohort study, are participants in the exposed and unexposed groups similar in all important respects except for the exposure?

In a case-control study, are cases and controls similar in all important respects except for the disease in question?Information biasIn a cohort study, is information about outcome obtained in the same way for those exposed and unexposed?In a case-control study, is information about exposure gathered in the same way for cases and controls?

ConfoundingCould the results be accounted for by the presence of a factor— e.g. age, smoking, sexual behaviour, diet—associated with both the exposure and the outcome (but not directly involved in the causal pathway)?

Slide24

Evidence?

Bradford Hill

criteria

for evidence of causation…Strength of association:

yes (brazil, french polynesia: RR>20)Consistency: yes (brazil, polynesia, reports from Colombia)Specificity: yes (specific clinical presentation)Temporality: yes (individual infection and regional Zika outbreak occured before individual microcephaly cases and regional

peaks

)Plausibility: yes (other teratogenic viruses exist, like CMV or rubella; Zika virus found in fetal brains; infecting neural progenitor

cells in vitro)Coherence: yes neurotropic virus (Guillain barré syndrom in adults, animal models); infecting neural progenitor cells in vitroAnalogy: yes other teratogenic virusesBiologic gradient: n/a (infection is a Yes/No exposure)Experiment: no experimental model availableRasmussen et al, NEJM 2016

Slide25

Other alternative designs

25

Slide26

First alert from

surveillance: excessive

number

of microcephaly cases in BrazilPassive recording of

events shows an abnormality…PAHO epidemiological report Brazil – Sept 2016

Slide27

First alert from

surveillance:

excess

number of microcephaly cases following Zika outbreak…

Johansson et al, NEJM 2016

Slide28

First alert from

surveillance:

excess

number of microcephaly cases following Zika outbreak…

There were Zika outbreaks well described before, but cases of microcephaly remained unnoticed, probably because these outbreaks occured in small-sized populations :- In French Polynesia, the 2013-2014 Zika outbreak allowed description of association between Zika virus infection and Guillain Barré syndrom. In light of reports from

Brazil

, a

mathematical modelling work retrospectively identified an excess of cases of microcephaly after the outbreak (8 observed vs 1 expected).Cao-

Lormeau et al, Lancet 2016; Cauchemez et al, Lancet 2016

Slide29

Case-reports

Reporting

one

well-described case, with molecular identification of the

sequence of Zika virus in the abnormal brain of the fœtus.Reporting one well described case, with identification IgG directed against Zika virus.

Slide30

Ecological correlation studies

Ecological

correlation

between low yellow fever coverage and high number of cases of microcephaly related to Zika virus infection…

Slide31

Mathematical modelling study

Retrospective

data on

Zika-virus epidemic in French Polynesia 2013-2014

Incidence rate of Zika infection by week Prevalence of infection before and after, by age group ( attack rate in the population)Birth rate in the populationNumber of microcephaly cases recorded and date ( in total 8 cases)No individual data available

on

mother

/child Zika infection statusDesign: mathematical model to explore

which hypothesis on the impact of Zika infection explains best the observed temporal distribution of these 8 cases.Hypothesis: p(t): no impact, impact if 1st trimester, if 1+2, anytime during

pregnancy… Pregnant women (susceptible)Pregnant women (

infected)Normal baby deliveredMicrocephaly casei(t)p(t)Zika outbreak ongoing in population

Cauchemez

et al, Lancet 2016

Slide32

Mathematical modelling study

The

observed

distribution of

microcephaly cases was found to be most compatible with an impact of Zika virus infection occuring during trimester 1, 1+2 or 1+2+3.If no association existed with Zika virus, the observed distribution was very different from what is expected (uniform incidence of microcephaly due to other causes).

Estimated

increase in risk of microcephaly due to Zika infection ranging between 20x to 50x depending on the period where infection by the virus is causing defects.

Cauchemez et al, Lancet 2016

Slide33

Key points The study question must be precisely identified (PICO)

Design should match the research question

Analysis should match the design

Interpretation should be appropriateStudy findings should be published fully and accurately Planning is vital – trial protocol Methodological input is valuable (essential) at every stage

33

Slide34

Quotes

“100% of all disasters are failures of design, not analysis. “

--

Ron Marks, Toronto, August 16, 1994

“There are only a handful of ways to do a study properly but a thousand ways to do it wrong.”-- L Sackett. Rational therapy in the neurosciences: the role of the randomized trial. Stroke, 1986 “To propose that poor design can be corrected by subtle[statistical] analysis techniques is contrary to good scientific thinking.” -- Stuart Pocock (Controlled Clinical Trials, p 58)“Issues of design always trump issues of analysis. “ -- GE Dallal, 1999 (it would be wasted effort to focus on the analysis of data from a study whose design was fatally flawed)