Since matching was done without replacement and by sequentially scanning through the trial data, the matching could depend on the initial order of the individuals in the trial dataset. To take this into account, matching was done based on ten distinct random orders of the trial dataset. To increase the sample size and hence statistical power, GW833972A multiple DCH Encainide hydrochloride participants were matched to each trial participant. After the first full scan of the trial data, a second iteration was started, etc. However, with an increasing number of iterations, the distances of the matches increase; hence there is a trade-off between sample size and matching quality. To decide how many DCH participants to match each trial participant, scree plots were inspected. Figure 2 shows, as an example, a scree plot, based on one of the random orders of the trial participants, when matching on protein E%, carbohydrate E%, GI, BMI and WC. Corresponding graphs based on other matching-combinations looked similar. After the third iteration, the distance-increase began to level off in all scree plots, so four iterations were used, i.e. four DCH participants were matched to each trial participant. The group of individuals selected from the initial observational cohort data chosen by matching is referred to as the modified observational data below. Figure 3 shows the matching-performance when matching on protein E%, carbohydrate E%, GI, BMI and WC, based on mean values across matches of the ten random orders of the trial dataset. For each match-variable, the values of the four DCH participants are plotted against the value of the matched trial participant. Corresponding plots based on other match-combinations looked similar; see Figures S1�CS4 in file S1. A hypothetical, perfect match would have followed the straight line of equality. As seen, it was not possible to get a very close match in the observational data of the greatest protein E% intakes of the trial participants. A similar, although much less prominent, pattern was observed regarding match on carbohydrate E% and GI. The match on BMI and WC was fair, even though deviations tended to increase with higher values. Figure 3 also distinguishes between the high and low protein group. Considerable variation in protein- and carbohydrate intake was present within the groups of trial participants randomized to high or low protein, and hence also among the participants of the modified observational data.