Theoretical investigation of the impact of different timing schemes in hypofractionated radiotherapy.

PURPOSE: This study compares the effectiveness of three fractionation schemes of equal fraction size, comprising five fractions of SBRT over 5 days, 10 days, or 15 days, respectively. METHOD: This comparative study is based on two tumor-control-probability (TCP) models that take into account tumor cell re-sensitization and repopulation during treatment; the Zaider-Minerbo-Stavreva (ZMS) and the Ruggieri-Nahum (RN) models. The ZMS model is further modified to include also re-sensitization according to the ß mechanism of the linear-quadratic (LQ) model of cell killing. The modified version of the ZMS model is verified through fitting to the experimental data set of Fisher and Moulder. The study applies an idea used in a plan ranking methodology developed for the case when the specific values of the model parameters are not known. RESULTS: The TCPs of the compared regimens are calculated for various values of the model parameters and for two different values of the dose per fraction. The TCPs are presented as 2-D functions of two of the model parameters for each model correspondingly. The differences between the TCPs of each of the prolonged regimens and the TCP of the every week day regimen are also calculated for each model. CONCLUSIONS: Both models predict that the prolonged regimens are superior in terms of TCP to the every week-day one for most of the studied cases; however this is shown to exist to a different degree by the two models. It is shown again to a different degree that reversed situations where the every week day schedule is better than the prolonged regimens are also possible. It is concluded that a 30% TCP difference observed in a clinical study in favor of the fifteen-day regimen is theoretically possible. However, the fifteen-day regimen is outperformed in terms of TCP by the every week day regimen in more cases than the regimen lasting ten days. Therefore the choice of a prolongation in time must be made with care.

Predicting healthcare trajectories from medical records: A deep learning approach.

Personalized predictive medicine necessitates the modeling of patient illness and care processes, which inherently have long-term temporal dependencies. Healthcare observations, stored in electronic medical records are episodic and irregular in time. We introduce DeepCare, an end-to-end deep dynamic neural network that reads medical records, stores previous illness history, infers current illness states and predicts future medical outcomes. At the data level, DeepCare represents care episodes as vectors and models patient health state trajectories by the memory of historical records. Built on Long Short-Term Memory (LSTM), DeepCare introduces methods to handle irregularly timed events by moderating the forgetting and consolidation of memory. DeepCare also explicitly models medical interventions that change the course of illness and shape future medical risk. Moving up to the health state level, historical and present health states are then aggregated through multiscale temporal pooling, before passing through a neural network that estimates future outcomes. We demonstrate the efficacy of DeepCare for disease progression modeling, intervention recommendation, and future risk prediction. On two important cohorts with heavy social and economic burden - diabetes and mental health - the results show improved prediction accuracy.