Table 1.
Methods for analysis n-of-1 trials.
Table 2.
Degrees of freedom for a design with the block structure Patient/Cycle.
The second column gives the degrees of freedom for treatment structure Treatment and the third with Treatment. Patient added. The case of 12 patients and 3 cycles is illustrated.
Table 3.
Degrees of freedom for a design with the block structure Patient/Cycle.
The second column gives the degrees of freedom for treatment structure Treatment and the third with Treatment. Patient added. The general case is illustrated.
Table 4.
Degrees of freedom for a design with the block structure Patient*Period and the treatment structure Treatment.
The case of 12 patients and 6 periods.
Table 5.
Degrees of freedom for a design with the block structure Patient*Period and the treatment structure Treatment.
The case of n patients and m periods.
Fig 1.
Comparison of true and estimated variances for the matched pairs approach (see text for explanation).
Fig 2.
Comparison of estimated variances for the summary measures approach and the mixed model with the true variance.
Table 6.
Simulated data from a trial in asthma.
The data are for 12 patients treated in 3 cycles. The data are arranged in columns by treatment given. For each patient the first row represents the period in which the treatment was given and the second the result in ml of FEV1. Values in italics and underlined are those which are removed to create an unbalanced set.
Table 7.
Naïve per patient treatment estimates and estimated standard errors for the 12 patients whose data are given in Table 6.
The columns headed k give the number of observations per patient. See text for explanation.
Table 8.
Shrunk estimates and standard errors for a meta-analysis of the balanced case and unbalanced cases of Table 7 using the metafor package.
Fig 3.
Shrunk and naïve estimates for 12 patients in the unbalanced case.
The solid diagonal line is a line of equality. It can be seen that the estimates for individual patients are strongly shrunk and the dotted line gives the line of shrinkage for patients 1 to 10 who all have the same amount of information. The value for patient 11 is shrunk more strongly than for patients 1–10 because data from one cycle are missing. For patient 12 data from two cycles are missing and shrinkage is even stronger.