Relapse Trend Score

The application of a symbolic series-based approach to compute the relapse trend score is explained in detail in the Supplementary Material. Briefly, this method captures the relapse onset temporal pattern for each individual based on the time gap between relapses and converts it into a score between 0 to 1 that describes the trend observed in these time gaps. An individual with a history of n relapses (n ≥ 4) has n-1 time units between the relapses (Figure 1). This series of time units, when iteratively grouped into three consecutive time units—with a shift in one time unit each iteration—will result in n-2 trend units. Each trend unit is a measure of direction of change between consecutive three time units and takes a score of 1 if consecutive time units form a predictable pattern, that is, are increasing or decreasing or remaining the same, and takes a score of 0 otherwise, in which case they are non-predictable or random patterns. The sum of these n-2 trend unit scores is then normalized within each cohort group based on the number of relapses. Thus, the temporal pattern score for an individual with n relapses is 1 when all n-2 trend scores are 1 and is 0 when none of the n-2 trend units scored is a 1, and anything in between will be a score between 0 to 1. A lower score suggests that relapse onset or time between hospitalizations is random, and a higher score suggests predictability over time. Figure 2 illustrates trend scores for two individuals with 6 relapses, one with a predictable trend and the other with unpredictable relapses. The above methodology is explained in more detail in the Supplementary Material.

Schematic of a MH patient trajectory indicating hospitalizations and relapses.

Relapse time trajectory of a patient with five relapses where the time to relapse is progressively decreasing between each hospitalization (left). Relapse time trajectory of a patient with five relapses where the relapse times are irregular and do not follow a trend (right).