Merridown analysis


  • The final predictive model for CIMT Max included SBP, Hba1c and Homocystein, as well as the interaction between SPB and Hba1c.

  • SBP, Hba1c and Homocystein are all affected by Statin use and the use of BP medication

  • SBP is affected by Hba1c and Homosystein

  • Apoliprotein A1 is a significant predictor of Hba1c

  • Apoliprotein A1 shares significant relationships with HDL and Statin use.

Analysis of complete records/missing data

Across all patient records, there were no complete records for data collected between 2008 and 2013. To investigate the missing data structure, an indicator matrix was constructed with the same dimension as the dataset. Each entry was coded as a 0 or 1, depending on whether an observation was recorded. A hierarchical clustering algorithm was then applied over patient records, to determine which variables were most similar with respect to information recorded (Figure \ref{fig:dendrogram}). This gave insight into which variables were commonly collected together, for example, as part of the same blood test.

\label{fig:dendrogram}Final dendrogram detailing most likely variables groupings. The clustering of variables was used to fit smaller models for examining effects on CIMT.

Univariate repeated measures ANOVA

To determine a subset of likely covariates that influenced CIMT, a linear mixed model with subject random effects was fitted to each explanatory variable, using CIMT Max as response variable. A covariate was selected as a candidate for the final model if its p-value was less than or equal to 0.05. The results of this analysis are summarised in Figure \ref{fig:CIMTMaxANOVA}. Based on these results, 6 variables were were associated with CIMT_Max: ApoliA1, Hba1c, Homocystein, SBP, Statins (Yes/No) and Blood thinning medication (Yes/No).