Testing for a Sweet Spot in Randomized Trials

Figures

• 

  Figure 1. Relative benefit and continuous outcome. Hypothetical data from a trial where primary outcome is a continuous variable that gauges clinical effectiveness. The x-axis denotes disease severity measured at baseline. The y-axis denotes relative change from baseline following treatment. (A) Uniform relative benefit at all levels of disease severity. (B) Potential sweet spot with greater relative benefit at middle range and less benefit at extremes of disease severity. An example could be a diet intervention for overweight patients where the average participant drops 0.1 pounds per day, yet those at extremes of baseline weight have smaller relative changes.

• 

  Figure 2. Cumulative benefit and continuous outcome. Hypothetical data from a trial where primary outcome is a continuous variable that gauges clinical effectiveness. The x-axis denotes disease severity measured at baseline. The y-axis denotes cumulative change in outcome from baseline following treatment. (A) Uniform relative benefit at all levels of disease severity. (B) Potential sweet spot with greater relative benefit at middle range and less benefit at extremes of disease severity. An example could be a diet intervention for overweight patients where the total cohort drops 80 pounds per day, yet those at extremes of baseline weight have smaller relative changes.

• 

  Figure 3. Survival in quintiles of disease severity. Analysis of survival benefit stratified by of five categories of disease severity. X-axis shows disease severity with least severe quintile on left and most severe quintile on right. Y-axis shows proportion of each group alive at study termination. Bars show observed data and circles show results from Gompertz model. Speckled pattern denotes defibrillator group, striped pattern denotes control group, and floating percentages indicate actual survival in each group. Lower square brackets provide count of triplets, individual patients, observed deaths, and estimated number-needed-to-treat (NNT) in each quintile. Results show higher probability of survival with defibrillator treatment for each quintile, accentuated increase in middle quintile, and close fit of Gompertz model with observed data in all quintiles.

• 

  Figure 4. Cumulative survival advantage for defibrillator patients. Histogram of cumulative survival advantage comparing defibrillator patients to control patients. The x-axis shows consecutive triplets of matched patients who are sequenced ordinally by increasing disease severity. The y-axis shows cumulative count of survival advantage for defibrillator patients. Display contains 829 bars for 829 matched triplets (1 defibrillator patient and 2 control patients in each triplet, all with similar disease severity). Results show survival advantage of defibrillator patients mostly explained by individuals with intermediate disease severity.