Farmacoepi Course Leiden 0210 Part 2

faculteit Farmaceutische Wetenschappen

PharmacoepidemiologyStudy designInformation bias & selection bias

Rob Heerdink


Dr Rob Heerdink

Pharmacoepidemiology & Pharmacotherapy

Utrecht Institute for Pharmaceutical Sciences

Universiteit UtrechtThe Netherlands

www.pharm.uu.nl/epithera



Drug development

discovery

Discovery & screening

Proof of Concept

first administration to man

registration& launch

approx. 10-12 years

10,000

Pre-clinicaldevelopment

15-30

Fase I/IIa

10-15Fase IIb/III

15

preclinicalclinical (I-III)

phase IV


Relevant questions in practice following registration• effect on hard endpoints

• long term (side)effects

• value compared to other drugs

• effect in populations that were not studied

• children

• elderly

• pregnant

• multiple pathology / drug use

• who benefits and who does not

• less frequently seen adverse effects


DOMAIN

Determinant(s) Endpoint(s)time

• yes / no comparison

• experimental or observational

• retrospective or prospective

Study design


Research designs

Follow-up (cohort)Prospective

Experimental (clinical trial)Observational

Retrospective

Case controlCase series


Hierarchy of designs

1. Meta-analysis of clinical trials2. Clinical trial3. Prospective follow-up (observational)4. Retrospective follow-up5. Case control6. Case series


Study design

Past Present FutureRetrospective Cohort Prospective Cohort

Case-Control (retrospective)

Cross-sectional


Clinical trials– Randomisation

• Aim is to create groups with similar prognosis

– Control group: comparison with placebo or active treatment

• Aim is to evaluate the pharmacological effect of the new compound (exclusion of placebo effect, natural course, measurement error)

– Blinded (single, double, triple)• Aim is to prevent subjectivity (in patients, researcher,

statistician) in the scoring of the prognostic factors and outcome and to prevent differential changes in behaviour (changes that might change prognosis) of the groups to be compared.


Source population

Randomisation

Index group Control group

Follow-up Follow-up

Outcome Outcome

In- and exclusion criteria

Method, blinding

Prognostically comparable

TreatmentDouble blind

Loss-to-follow-up

Blinded measurement of outcome


Pharmacoepidemiological designs•Descriptive methods (Signal detection, hypothesis generating).

Identifying previously unrecognised safety issues – Case reports, – Case series, – Cross-sectional study

•Analytical methods (quantifying + risk factors, hypothesis testing). Investigating possible hazards (hypothesis-testing in order to substantiate a causal association)

– Observational • Cohort studies, • Case-control studies, • Case-crossover studies

– Intervention • Experimental Clinical trial


Evaluation of therapy: golden standardRandomised Controlled Clinical Trial (RCT)

Randomise: why?

Controlgroup: why?

Blinding: why?

Goal:Only difference between treated and untreated group is the treatment


Experiments are often impossible

Ethical (e.g. smoking, birth defects)

Practical (e.g. rare adverse effects)

Non-experimental (observational) research

For example:

Do animals bite more often during full moon?


Do animals bite more during a full moon?

Bhattacharjee C et al. BMJ 2000;321:1559-61


Observational studies

Past Present FutureRetrospective Cohort Prospective Cohort

Case-Control (retrospective)

Cross-sectional


Case Report / Case series

Describes characteristic association in one / somepatient(s) between determinant en outcome

examples:

• serious liverdamage following use of XTC

• birth defects after use of Thalidomide (Softenon)

• etcetera, etcetera, etcetera


The Lancet, 1961


LETTER TO THE EDITOR

THALIDOMIDE AND CONGENITAL ABNORMALITIES

Sir, Congenital disorders are present in

approximately 1.5% of babies. In recent months

I have observed that the incidence of multiple

severe abnormalities in babies delivered of

women who were given the drug thalidomide

('Distaval') during pregnancy,as an anti-emetic

or as a sedative, to be almost 20%.

Have any of your readers seen similar

abnormalities in babies delivered of women

who have taken this drug during pregnancy?

McBride WG. The Lancet, December 16, 1961: page 1358


Dwarsdoorsnede onderzoek

beschrijft een karakteristieke relatie tussen determinant en uitkomst op 1 moment in de tijd

Causaliteit?



Voorbeeld dwardoorsnede onderzoek

Polymorphisms of the LEP- and LEPR gene and obesity in

patients using antipsychotic medication

Gregoor et al J Clin Psychopharmacol (2009)

Onderzoeksvraag: zijn polymorfismen in de LEPR

geassocieerd met hoger BMI in antipsychotica

gebruikers?

Studie opzet: dwarsdoorsnede onderzoek


Voorbeeld: LEPR onderzoek

Populatie: 200 antipsychotica gebruikers

Determinanten: LEPR Q223R en LEP promoter 2548G/A SNP polymorfismen

Uitkomst: BMI


N BMI>30

Males

QQ 30 6 (20%)

QR 73 16 (21%)

RR 31 8 (26%)

Females

QQ 17 12 (71%) **

QR 39 15 (39%)

QR 10 4 (40%)

** p<0.05

Voorbeeld: LEPR onderzoek


Observational Cohort

Group of individuals with common inclusion criteria is followed over time until an endpoint occurs.


Cohort study / Follow-up study

Study population

Exposed

Non-exposed

Disease +

Disease +

Disease -

Disease -


Voorbeeld cohort onderzoek

• Zijn patiënten die van specialité naar generiek antihypertensivum switchen minder therapietrouw?

• Blootstelling: switch• Uitkomst: therapitrouw


Switchers: 13,6% therapie-ontrouw

Niet switchers: 18,7% therapie-ontrouw

Van Wijk et al 2006


A cohort studyRR (myocardial

infarction)*

Untreated normotensive and hypertensive men 1.0 (reference)Treated hypertensive men DBP90 mmHg 3.8 (1.3-11.0)Treated hypertensive men DBP>90 mmHg 1.1 (0.5-2.6)

* adjusted for previous MI, CVA, IHD, IC, diabetes, SBP, duration of antihypertensive therapy, hypercholesterolemia, hypertriglyceridaemia, creatinin, obesity, use of cardiac glycosides, smoking.

Conclusion: In men treated for hypertension, DBP should not be reduced to lower than 90 mmHg

Merlo J, et al. BMJ 1996;313:457-61.


Another cohort study

RR (stroke)

Untreated “Candidates”* for treatment 1.0 (reference)

Treated Crude RR 0.49 (0.32-0.76)Adjusted RR* 0.61 (0.39-0.97)

** Adjusted for age, sex, diabetes, total cholesterol, BMI, smoking, history of CVD

* Candidates for treatment defined according to Dutch guidelines on treatment of hypertension taking into account multifactorial risk of cardiovascular disease

Klungel et al. Epidemiology 2001;12:339-344.


Frequency measures cohort study

(P1 personyears)

(P0 personyears)

A1

A0

Exposure

No Exposure

Disease

Disease

No disease

Time

Time

No disease (B1)

(B0)


Frequency measures

• Incidence– Cumulative incidence (CI)– Incidence rate (IR)


Risk disease + | exposure + = A1 / N1 = CI1

Risk disease + | exposure - = A0 / N0 = CI0


Disease No disease Total

Exposure + A1 B1 N1

- A0 B0 N0


Force of morbidity | exposure + = A1 / P1 = IC1

Force of morbidity | exposure - = A0 / P0 = IC0


Disease Personyears

Exposure + A1 P1

- A0 P0


Risk difference RD CI1 – CI0 IC1 - IC0

Relative Risk RR CI1 / CI0 IC1 / IC0

Attributable Risk AR (CI1 - CI0) / CI1 (IC1-IC0) / IC1

Relative Risk Reduction RRR 1 – RR

Number needed to treat NNT 1 / RD

Effect measures cohort study


Pill and Deep Venous Thrombosis

Risk no pill = 3.9 per 100 000 pyRisk pill gen. 2 = 10.3 per 100 000 pyRisk pill gen. 3 = 21.3 per 100 000 pyRR2/3 = 2.07

RV2/3 = 11.0 per 100 000 py

AR2/3 = 52%

‘NNH’2/3 = 9091 per yearLancet 1995; 346: 1582 - 1588


Analysis cohort study

• Survival analysis– Kaplan Meier (CI)– Cox regression (I)

• Proportional hazards model:• ln (Inc) =h0(t) + b1xX1 + b2xX2 + … + b1xXi

• Interpretation: eb1=HRx1 (adjusted for X2 .. Xi)


Kaplan Meier


Prospective vs. retrospective Cohort Studies

• Prospective Cohort Studies– Time consuming, expensive– More valid information on exposure– Measurements on potential confounders

• Retrospective Cohort Studies– Quick, cheap– Appropriate to examine outcome with long latency periods– Admission to exposure data– Difficult to obtain information of exposure– Risk of confounding


Selection of the Exposed Population

• Sample of the general population:– Geographically area, special age groups, birth cohorts

(Framingham Study)

• A group that is easy to identify:– Nurses health study

• Special population (often occupational epidemiology):– Rare and special exposure– Permits the evaluation of rare outcomes


Selection of the Comparison Population• Internal Control Group

– Exposed and non-exposed in the same Study population (Framingham study, Nurses health study)

• Minimise the differences between exposed and non-exposed

• External Control Group– Chosen in another group, another cohort (Occupational

epidemiology: Asbestosis vs. cotton workers)

• The General Population


Follow up study versus RCT

• Similarities– Use of same measures of frequency and association

(RR, RD, AR, RRR, NNT, NNH)– Use of same analytical techniques (“survival”

analysis: Kaplan Meier curves and Cox proportional hazard)

• Differences– Follow-up vs. RCT: no randomisation and no blinding

(outcome measurement sometimes blinded)

Follow-up studies more vulnerable to bias


Cohort study / Follow-up study

Study population

Exposed

Non-exposed

Disease +

Disease +

Disease -

Disease -


Case-control study

Study Population

Cases

Controls

Exposed

Non-exposed

Exposed

Non-exposed


Example case-control study

• What is the risk on breast cancer with the use of SSRI antidepressants?

• Cases: women with breastcancer• Controles: women with no breastcancer• Exposure: SSRIs


Coogan et al. Am J Epidemiol 2005


Meaning of odds and odds ratio ?

• Term from English gambling “4 to 1”

• Odds means chance of succes (Ps) / chance of no success (Pns)

• Pns = 1-Ps

• Odds = Ps / 1-Ps

• Ps = 1/5 =0,2 => odds = 0,2 / 0,8 = 0.25 (4 to 1)

• In epidemiology ‘Exposure odds’ of interest


Case Control

Exposure + a b

- c d

a+c b+d

Exposure odds among cases = a/c

Exposure odds among controls = b/d

Exposure Odds ratio (OR) = (a/c) / (b/d)

= (a*d) / (b*c) RR

Calculation OR


Rofecoxib and risk of MI

Nested Case Control design:• 9218 MI cases of whom 93 used of rofecoxib < 3 months ago• 86349 controls, of whom 634 used of rofecoxib < 3 months ago

MI control

Rofecoxib + 93 634

- 9125 85715

OR (MI)= 93x85715 / 634x9125 = 1.38BMJ 2005;330:1366


Selection of cases

• Establish strict diagnostic criteria for the outcome: Examples:– Type 1 diabetes in children: severe

symptoms, very high BG, marked glycosuria, and ketonuria.

– Type 2 diabetes: few if any symptoms, Slightly elevated BG, diagnosis “complicated”.


Selection of cases

• Population-based cases: include all subjects or a random sample of all subjects with the disease at a single point or during a given period of time in the defined population:– Disease registers

• Hospital-based cases:All patients in a hospital department at a given time


Selection of Controls

Principles of Control Selection:• Study base:

– Controls can be used to characterise the distribution of exposure

• Comparable-accuracy– Equal reliability in the information obtained from cases and

controls no systematic misclassification

• Overcome confounding– Elimination of confounding through control selection

matching or stratified sampling


Selection of Controls

• General population controls:– registries, households, telephone sampling– costly and time consuming– recall bias– eventually high non-response rate

• Hospitalised controls:– Patients at the same hospital as the cases– Easy to identify– Less recall bias– Higher response rate


Ascertainment of outcome and exposure status

• External sources:– Death certificates, disease registries,

Hospital and physicians records etc.

• Internal sources: – Questionnaires and interviews, information

from a surrogate (spouses or mother of children), biological sampling( e.g. antibody)


Data CollectionExternal Data

SourcesInternal Data

Sources

Exposure Hospital records, employers

Questionnaires, physical examinations, and/or blood tests, other diagnostic tests

Event Disease registries, death certificates, physician and hospital records

Questionnaires, physical examinations, and/or blood tests, other diagnostic tests

Confounder Hospital records registries

Questionnaires, physical examinations


Strengths in Cohort vs. Case-control?

Cohort study• Rare exposure• Examine multiple effects

of a single exposure• Minimizes bias in the in

exposure determination• Direct measurements of

incidence of the disease

Case-control study• Quick, inexpensive• Well-suited to the evaluation

of diseases with long latency period

• Rare diseases• Examine multiple etiologic

factors for a single disease


Limitations in Cohort vs. Case-control?

Cohort study• Not rare diseases• Prospective: Expensive

and time consuming• Retrospective: in

adequate records• Validity can be affected

by losses to follow-up

Case-control study• Not rare exposure• Incidence rates cannot be

estimated unless the study is population based

• Selection Bias and recall bias


Validiteit +

Precisie -


Validiteit -

Precisie +


Validiteit +

Precisie +


Validiteit -

Precisie -


Precisie en validiteit (vertekende resultaten)

• Toevallige fouten: ‘random error’ (precisie)

steekproef

• Systematische fouten: ‘systematic error’ (validiteit)

bias en confounding


Power and precision


Sample size in case-control study of OC use and MI among women

Assuming proportion of current OC use of 10%;

•Power: 1- (type II error) = 80%

•Precision: (type I error) = 5%


Sample size requirement in RCT

baseline risk number required side effect in each groupin control group

50% 1425% 7710% 2665% 5821% (liver dysfunction) 3,1040.1% (hepatitis) 31,4830.01% (cholestatic jaundice) 315,268


The likelihood of observing an adverse drug reaction in 2,000 patients

Threshold for ADRProbability

1 / 500 0.98

(Lymphoma From Azathioprine)

1 / 1,00 0.86(Eye Damage From Practolol)

1 / 10,000 0.18(Anaphylaxis From Penicillin)

1 / 50,000 0.04 (Aplastic Anemia From Chloramphenicol)

Lembit Rägo, WHO Upsala


Validiteit van onderzoek

• Komen de bevindingen uit het onderzoek overeen met de werkelijkheid?

• Externe en interne validiteit

Externe validiteit: representativiteit en non-response Interne validiteit: selectie bias, informatiebias en

confounding bias


Soorten van selectie- en informatiebias

• Channeling bias• Referal bias• Diagnostic bias

• Observer bias• Recall bias• Response bias

selectie informatie


Information bias (follow-up)

• Question: do patients who are treated for Hodgkin’s disease have a higher chance of a second tumor ?

– Compared incidence of cancer in a group of treated Hodgkin patiënts with incidence of cancer in general population (matching on age and gender)

Information bias (if RR = 1 probably no bias; if RR>1 cave bias)


Selection bias (cohort)• Observational cohort study

RR (myocardial infarction)*

Untreated normotensive and hypertensive men 1.0 (reference)Treated hypertensive men DBP90 mmHg 3.8 (1.3-11.0)Treated hypertensive men DBP>90 mmHg 1.1 (0.5-2.6)

* adjusted for previous MI, CVA, IHD, IC, diabetes, SBP, duration of antihypertensive therapy, hypercholesterolemia, hypertriglyceridaemia, creatinin, obesity, use of cardiac glycosides, smoking.

Conclusion: In men treated for hypertension, DBP should not be reduced to lower than 90 mmHg

Merlo J, et al. BMJ 1996;313:457-61.


Selection bias (cohort)

• Observational cohort study in the NetherlandsRR (stroke)

Untreated “Candidates”* for treatment 1.0 (reference)

Treated Crude RR 0.49 (0.32-0.76)Adjusted RR* 0.61 (0.39-0.97)

** Adjusted for age, sex, diabetes, total cholesterol, BMI, smoking, history of CVD

* Candidates for treatment defined according to Dutch guidelines on treatment of hypertension taking into account multifactorial risk of cardiovascular disease

Klungel et al. Epidemiology 2001;12:339-344.


Case control

Time

Time

Exposure

No exposure

Exposure

No exposure

Case

(disease)

Control

(no disease)


Selection of controls

Smoking and leukemia cases: patients with leukemia controls: patients in hospital for other disease

Patients hospitalized for CVD as control ?

Patients with history of CVD, not hospitalized for CVD

Smoking => CVD


Information bias (case-control)

Drugs and congenital malformation

Cases : children with congenital malformation

Controls: children without malformation

Exposure: use of drugs


Selection bias (case-control) I

Oral contraceptives and DVT

It was published (prospective follow up) in the 60’s after introduction of OAC that OAC increase the risk for DVT 3 times

New study planned: case control design Cases? Controls?


Selection bias (case-control) II

Women with pain in leg Physician asks about use of the pil If use of pil, refer to hospital, otherwise not.

RR DVT OC use

Solution ?


Methoden om voor vertekende resulaten te corrigeren

• Restrictie / StratificatieMantel Haenszel

• Matchen1:1 of proportioneel

• Mathematische modellenverschillende regressietechnieken (bij patiënt controle logische regressie; bij follow-up Cox of Poisson regressie)


Risk assessment





Registration of a drug is only the beginning of safety research

email: [email protected]: @robheerdinkwww.slideshare.net/robheerdink

Farmacoepi Course Leiden 0210 Part 2

Health & Medicine

casecontrol studies

case report case series

retrospective followup

observational studies

prospective study design

disease disease

preclinical clinical

preclinical development