Pharmacoepidemiologic study design in Europe: The example of insuline glargine and cancer risk

1
Pharmacoepidemiologic study design in Europe The
example of insuline glargine and cancer risk
Study design 4 examples

• Niklas Hammar
• AstraZeneca RD, Södertälje
• Department of Epidemiology, Institute of
  Environmental Medicine, Karolinska institutet

2
Conflicts of intererst

• Regional Manager Epidemiology Europe, AstraZeneca
  RD
• Assoc prof, adjunct lecturer Karolinska
  institutet
• Board member Swedish Society of
  Pharmacoepidemiology
• This presentation represents only my personal
  view

3
Pharmacoepidemiologic information in Europe
heterogenous

• Good conditions in some countries (UK, NL, Nordic
  countries)
• Differences in how databases are formed e.g.
  GP-based in UK/NL (GPRD, THIN, PHARMO),
  population based national health registers in
  Nordic countries
• Differences in registered information, coding and
  possibilities of follow-up
• Differences in possibilities to access data /
  collaborations
• Varied pharmacoepidemiologic experience and
  competence

4
Pharmacoepidemiologic study design in Europe

• Purpose of this seminar
• Illustrate conditions in different countries and
  of different approaches to a post marketing
  safety study
• Discuss comparative advantages and limitations
  using a recent example mainly from a
  methodological perspective

5
Example Insulin glargine and cancer risk
(Diabetologia 2009)

• Four pharmacoepidemiologic studies from three
  countries (Germany, Sweden, UK and Scotland)
  using data from four different kinds of sources
• Insurance claims data (Germany), national health
  registers (Sweden), GP-based data (THIN UK),
  diabetes register (Scotland)

6
Data sources
Germany Information on all insurants of Germanys largest health insurance fund representing 17.9 million people
Sweden 7 national registers were linked using the unique personal id number
Scotland Scottish Care Information-Diabetes Collaboration clinical database covering majority diagnosed diabetes in Scotland
UK The Health Information Network (THIN) created in 2002 includes data from about 300 UK general practices
7
Study population
Germany Patients gt18 y without known malignant disease who received first-time insulin therapy for DM exclusively with human insulin or with only one type of insulin analog (n127,031)
Sweden Subjects 35-84 y old at the end of 2005 who had at least one prescription dispensed for insulin 1/7 31/12 2005. Previous malignancy checked for (n114,841)
Scotland Patients with diabetes who received an insulin prescription during Jan 1 2002 Dec 31 2005 (n36,254) Fixed cohort Pat receiving any type of insulin July Oct 2003 Incident cohort T2DM pat starting insulin therapy first time Jan 2002 Dec 2005
UK Patients with a diagnosis of diabetes after 40 y of age who had received ?6 sequential prescriptions of oral hypoglyceamic agents and without other potential causes of secondary diabetes (n85,200) Insulin cohort those previously treated with OHAs who had been newly initiated on any insulin-based regimen
8
Exposure assessment
Germany Data from primary care prescriptions for insulin billed to the AOK including insulin type and dose
Sweden Information from the National Prescribed Drug Register on prescriptions dispensed between July 1 Dec 31 2005
Scotland Information from primary health care systems going into the SCI-DC name of drug and date of prescription
UK THIN includes information on prescribed medications
9
Insulin glargine exposure definition
Germany Insulin-naïve subjects (not received any prescription for human insulin or insulin analogues within 1 year prior to inclusion) exposed to glargine and no other insulin type. Only patients with at least one prescription every 6 months during follow-up. Mean daily dose was estimated as cumulative dose during follow-up divided by person time
Sweden At least one prescription for insulin glargine and not for any other type of insulin or insulin glargine and another type of insulin during July 1 Dec 31 2005
Scotland Fixed cohort Patients receiving insulin glargine July 1 Oct 31 2003 Incident insulin cohort Patients starting insulin therapy for the first time Jan 1 2002 Dec 31 2005 glargine during first 4 months of use
UK Patients previously treated with OHAs newly initiated on insulin glargine with no other concomitant insulin
10
Outcome
Germany Primary a malignant neoplasm Secondary all-cause mortality
Sweden Cancer All malignant tumours, all malignant tumours and in situ tumours, breast cancer (women), prostate cancer, gastrointestinal cancer Death from any cause, myocardial infarction
Scotland Incidence of all cancers and cancers at specific sites (breast, colon, prostate, pancreas, lung)
UK Primary first solid tumour cancer Secondary one of four solid tumour cancers recorded as the first cancer following treatment change in people with no prior cancer breast, pancreatic, colorectal or prostate cancer
11
Follow-up
Germany Jan 1 1998 June 30 2005 Mean 1.63 y Data from hospital claims billed to AOK (ICD-10 coded) survival status of AOK members
Sweden Jan 1 2006 Dec 31 2007 National Cancer Register (ICD-10 code), Patient register, Cause of Death Register
Scotland Jan 1 2002 Dec 31 2005 Scottish Cancer Registry, Death register of General Registrars Office for Scotland
UK 2000 -? Mean 2.42 y THIN contains information on past and current medical diagnoses
12
Confounders
Germany Age, sex, hospital stay (no. and duration), state, start of therapy, other medication, dose
Sweden Age, sex For 50-75 BMI, smoking, age at onset of DM, CVD, age at birth of first child (breast ca)
Scotland Prior cancer, type of DM, calendar year, metformin, sulfonylurea, other OADs, diabetes duration, HbA1c, DBP, SBP, deprivation, smoking, BMI (not available for all)
UK Age, sex, smoking, prior cancer, (HbA1c, diabetes duration, weight)
13
Main results Users of insulin glargine vs
reference group
All cancer Breast cancer All cause mortality
Germany Ref a 0.86 (0.36-1.09) Not studied 0.73 (0.70-0-77)
Sweden Ref b 1.07 (0.91-1.25) 1.99 (1.31-3.03) 0.83 (0.71-0.96) (women)
Scotland Ref c 1.02 (0.77-1.36) 1.49 (0.79-2.83) Not studied
UK Ref d 0.81 (0.59-1.11) 0.86 (0.42-1.75) Not studied
a) glargine alone vs. human insulin, b) glargine
alone vs. non-glargine insulin, c) glargine vs.
non-glargine insulin d) glargine alone vs. human
basal insulin
14
Main results Users of insulin glargine alone vs
reference group (a-b).
All cancer Breast cancer All cause mortality
Germany Ref a Dose adj and dose stratified 1.18 (1.08-1.28) 1.31 (1.20-1.42) for 50 IU (dose-dependent) Not studied 0.96 (0.92-1.01) 1.20 (1.11-1.30) for 50 IU
Sweden Ref b multivar adj 1.06 (0.90-1.25) 1.97 (1.30-3.00) 0.83 (0.71-0.96) (women)
Scotland Ref b Fixed cohort Incident cohort 1.55 (1.01-2.37) 0.87 (0.63-1.21) 3.39 (1.46-7.85) 1.47 n.s. Not studied
UK Ref b 0.81 (0.59-1.11) 0.86 (0.42-1.75) Not studied
a) human insulin, b) non-glargine insuline
15
Number of exposed and number of cases in users of
insulin glargine alone
Exposed All cancer Breast cancer
Germany 23,855 669 Not studied
Sweden 5,970 149 25
Scotland 447 22 6
UK 2,286 106 ? 10
16
Inconclusive results

• Editorial comments
• ..provide an excellent demonstration of the
  problems and pitfalls of observational studies
• A prospective clinical trial would be the best
  way of resolving the issue but would be
  unfeasible, arguably unethical and too slow to
  perform
• An additional approach would be a much larger
  pharmacoepidemiological analysis, jointly
  designed and conducted by indsutry and
  representatives of the scientific associations
  and independently analysed

17
Questions for group work

• Identify some strengths and limitations of the
  design in each of the four studies?
• What differences in study design or analyses may
  have contributed to differences in results and
  conclusions?
• What do you think about chance or confounding by
  indication as explanations for the breast cancer
  findings in Sweden and Scotland?
• Could a large European pharmacoepidemiologic
  study be designed to answer the question? Suggest
  important design components for this to work.
  Could/should it be a multi-centre study? Do we
  need to wait a few years more?
• How can we in general improve conditions for
  pharmacoepidemiologic studies in Europe?

18
Strengths and limitations
Strenghts Limitations
Germany Large study population Detailed exposure (incl dose) and confounders, follow-up on cancer and mortality Turnover, representativity, only total cancer and mortality, follow-up, data quality, short follow-up time, T2 vs T1 DM?
Sweden National registers of high quality, representativity, complete follow-up, diagnostic quality, PIN (linkage) Limited data on exposure and confounders for most subjects, short follow-up time, prevalent users of insulin
Scotland Good information on DM, detailed data on exposure and confounders, representative for Scotland, follow-up on cancer and mortality No data on insulin dose, fairly small population, short follow-up time, linkages?
UK Detailed data on exposure and confounders, probably representative for UK, T2DM No data on insulin dose, short follow-up time, events outside GP setting? Cancer identification and diagnosis?
19
Factors of possible importance for differences in
results and conclusions
DM type New users of insulin Glargine only Incident cancer Multivar adjustm Resid conf
Germany No ? Yes Yes Yes, 3y washout No (and yes?) Yes?
Sweden Yes No Yes and no Yes, since 1958 No and yes (50) Not likely?
Scotland Yes Yes and no Yes and no No Yes Likely?
UK T2DM Yes Yes Yes and no Yes Not strong?
20
Criticism of Hemkens et al

• Pocock Smeeth Lancet
• Unconventional (flawed) analysis adjusting for
  insulin dosage allocation to treatment groups
  must be determined before follow-up starts.
  Insulin-dose was calculated as the mean during
  follow-up then included as baseline covariate
• If a patient changed treatment or was ever on
  combined treatment they were removed from the
  analylisis. Any malignancy-free follow-up time
  before the change was excluded
• Higher dose will be linked to more severe disease
  and thus greater risk of cancer (confounding)
• Reverse causality insulin requierment influenced
  by undetected cancer
• Nagel JM et al Diabetiologia
• Numbers of prevalent vs incident insulin users
  unreasonable 3 of all pop have first
  prescription same date as cancer diagnosis
• Competing risks, confounding not appropriately
  discussed and sensitivity analyses called for
• Lack of comparability Glargine group mainly T2
  DM with higher risk of cancer ignorance of
  intrinsic insulin production in dose comparison