gms | German Medical Science

Introduction: In individualized treatment, a well-known strategy concentrates on identifying patient characteristics which predict response to a drug. Factors of this kind may be difficult to establish in some disease areas. In such cases, to achieve individualized treatment, the treatment schedule can be adapted at each monitoring time point in accordance to the observed response (individualized or as needed treatment). The treatment time points have to be taken into account as random variables. In addition, few patients only might share the same treatment pattern and the associated expected response profile. For instance, let us consider an ongoing 12 months clinical trial to evaluate two regimens of a treatment of the eye disease pathologic myopia. Following initial treatment, at each monthly visit the decision whether to retreat or not, is based on the patient's response. The number of treatments for a patient may vary between 2 and 12. Such individualized treatment patterns and response profiles over time pose challenges for modeling, parameter estimation and statistical inference, and for the interpretation of treatment effects.

Material and Methods: Following the approach in section 6.2 in [1], we propose a model which addresses these issues for the evaluation of adverse events, involving a piecewise-constant hazard rate and time-dependent covariates to reflect the diverse treatment patterns with varying treatment exposure. We present a maximum likelihood approach for estimation and inference.

Results: By modeling the safety risk under diverse treatment patterns, and by using maximum likelihood methodology for estimation and inference, it is possible to address targeted questions, e.g. to compare the safety risk under high and low exposure to treatment.

Discussion: By means of the proposed method, targeted questions can be answered, e.g. whether higher exposure to treatment is associated with a higher safety risk. This is useful for treatment decisions in clinical practice.

Topics for future research include extensions of the model, e.g. reflecting refined exposure-response relationships in the model.