# Feasibility of aligning creatine kinase MB activity and mass data in multicentre trials using generalized additive modelling

## TL;DR

This study shows that generalized additive models can accurately predict CK-MB mass from activity data, helping to align results across different centers in heart disease trials.

## Contribution

The novel contribution is the use of GAMs to align CK-MB activity and mass data across multicenter trials.

## Key findings

- GAMs predicted CK-MB mass with high accuracy (r2 = 0.981) using activity, sex, and sampling time.
- Bland–Altman analysis showed narrow agreement limits and no significant bias.
- Linear regression had lower accuracy and deviated at low CK-MB levels.

## Abstract

Elevated serum creatine kinase isoenzyme MB (CK-MB) levels indicate myocardial ischaemia and periprocedural myocardial injury during treatment of heart diseases. We established a method to predict CK-MB mass from activity data based on a prospective pilot study in order to simplify multicentre trials.

38 elective cardiac surgery patients without acute myocardial ischaemia and terminal renal failure were recruited. CK-MB mass and activity were determined in venous blood samples drawn preoperatively, postoperatively, 6 h post-op, and 12 h post-op. Linear regression and generalized additive models (GAMs) were applied to describe the relationship of mass and activity. Influences of demographic and perioperative factors on the fit of GAMs was evaluated. The agreement of predicted and measured CK-MB masses was assessed by Bland–Altman analyses.

Linear regression provided an acceptable overall fit (r2 = 0.834) but showed deviances at low CK-MB levels. GAMs did not benefit from the inclusion of age, body mass index and surgical times. The minimal adequate model predicted CK-MB masses from activities, sex and sampling time with an r2 of 0.981. Bland–Altman analyses confirmed narrow limits of agreement (spread: 8.87 µg/l) and the absence of fixed (P = 0.41) and proportional (P = 0.21) biases.

GAM-based modelling of CK-MB data in a representative patient cohort allowed to predict CK-MB masses from activities, sex and sampling time. This approach simplifies the integration of study centres with incompatible CK-MB data into multicentre trials in order to facilitate inclusion of CK-MB levels in statistical models.

Creatine kinase (CK; EC 2.7.3.2) is a dimer of M and B subunits.