Data Sources & Methods

Data Sources

The APCSC is a collaborative project collecting individual participant data from prospective studies in the Asia-Pacific region.

Data is supplied from 44 separate cohorts in 8 countries of the Asia-Pacific region. Some of these cohorts were initiated more than 30 years ago, while others are relatively new.

Methods

Study Eligibility

Studies are potentially eligible for inclusion in this project if they satisfy the following criteria:

  • A study population from the Asia-Pacific region
  • Prospective cohort study design
  • At least 5000 person-years of follow-up recorded or planned
  • Date of birth (or age), gender, and blood pressure recorded at baseline
  • Date of death or the age at death recorded during follow-up

Objective

To combine the data from the constituent cohorts appropriately, so as to precisely estimate relative risks for cardiovascular and other diseases for all major cardiovascular risk factors and standard socio-demographic measures.

Analysis Methods

Data were checked extensively for potential errors, which were queried with the relevant study center. Both categorical and continuous measures were standardized to common scales or conventions, using published criteria wherever possible and time-dependant Cox proportional hazard models were used to determine hazard ratios and 95% confidence intervals for the outcome associated with the risk factors of interest. The proportional hazard (PH) assumption for each variable was checked by the use of visual methods. The selection of variables to be included in the model were performed by comparing models that included gender, age, study and secular time, as potential major effect modifiers for the risk factors of interest. The associations between the continuous risk factors and the outcomes were adjusted for regression dilution bias, where the dilution coefficients were estimated using:

  • the non-parametric method of Peto and MacMahon; and
  • a mixed linear model that allows for different lengths of time intervals between repeated measurements and for a variance/covariance matrix with between-studies heterogeneity and within-subject autocorrelation.

Methodology Conclusions

  • Studies and sex do not satisfy the PH assumption; therefore we adopted a model that stratifies by these variables.
  • Age-at-risk is likely to be more sensitive than baseline age as a confounder.
  • Contrary to what was expected, secular time is not a confounder for any of the risk factors of interest. The final model includes, other than the risk factor investigated, age as continuous and time-dependant variable and sex and study as stratification variables.
  • A sex-by-risk factor interaction term was added to the model to derive a sex-specific hazard ratio: similarly a region by risk factor interaction was added to obtain the hazard ratios specific for Asian and Australasian populations.
  • Due to the error dilution bias, the effect of the risk factors was remarkably underestimated. For instance, the attenuation factor was 1.9 for systolic blood pressure and 2.3 for cholesterol. The results obtained from the Peto-MacMahon method and from the regression with the mixed models were very similar.

Data Ownership

The APCSC makes no claim of ownership on any of the data submitted to the Collaboration. The data from each study remain the sole property of the principle investigators of that study and will not be used for any presentation or publication without the consent of the collaborating investigators from the studies involved. Any data queries that may arise are discussed and resolved in confidence with the responsible collaborating investigator. The data provided for inclusion in the APCSC are held in strict confidence by the Secretariat.