Send Correspondence to journal:
Re: Donepezil treatment of patients with MCI. A 48-week randomized, placebo-controlled trial

I read with some concern the report by Doody and colleagues [1] who describe results of a randomized controlled trial of donepezil in patients with mild cognitive impairment (MCI). The drug was generally well-tolerated but the active treatment group experienced greater and earlier drop-out than those assigned to placebo, mostly due to gastrointestinal side effects.

The investigators confronted the common problem of dealing with missing data that were disproportionate across the comparison groups. Their chosen approach was the last observation carried forward (LOCF) method. In this method, the last recorded values prior to drop-out are retained in the analysis at every time point subsequent to the last actual observation.

In progressive degenerative diseases, when the active treatment causes premature drop-out, the LOCF technique is biased in favor of the active treatment. Other things being equal, earlier observations should show better cognitive performance than later observations. In other words, the LOCF method stacks the deck in favor of the drug.

To illustrate, imagine a comparison of two groups of five people. The “active” treatment has no effect other than causing early drop-out. Everyone starts at a score of 10 points and declines 1 point every 2 months. No one in the placebo-assigned group drops out. In the “active” treatment group, one individual drops out at month 3, 5, 7, and 9, leaving only one person to complete the study period. If we use LOCF it appears at 48 weeks that the “active” treatment is 2.8 points better on average than the placebo treatment (Figure).

Approaches to missing data have been an important area of research in statistics and biostatistics over the past several decades. It is disappointing that papers are still published using biased techniques such as LOCF.

Reference

1. Doody RS, Ferris SH, Salloway S, et al. Donepezil treatment of patients with MCI. A 48-week randomized, placebo-controlled trial. Neurology 2009;72:1555-1561.

Figure

Figure. Effect of the last observation carried forward (LOCF) method for missing data in a hypothetical trial with no difference between “active” treatment and placebo other than earlier drop-out in the “active” treatment arm. The placebo group shows a decline from 10 points to 4 points, while the “active” treatment group appears to show a smaller decline, from 10 points to 6.8 points, even though the only impact of “active” treatment was causing drop-out.

Send Correspondence to journal:
Re: Reply from the author

Dr. Crane correctly asserts that we utilized the LOCF for imputation of missing data in our trial. [1] Although there are concerns with every proposed method for dealing with missing data in clinical trials, an intention-to-treat analysis using LOCF was the common method used in anti-dementia trials when the statistical analysis plan for this study was written and reviewed by the FDA.

We reported a small and significant difference in Alzheimer Disease Assessment Scale–cognitive subscale (ADAS-Cog) scores favoring donepezil, but stated that the dual primary outcome for efficacy was not reached. Two previous MCI trials reported drug placebo differences favoring donepezil on the ADAS-Cog at 6 months in one study and up to 36 months in another. [2,3] This would support the possibility that this difference may be "real." However, our study remains negative because the psychometric outcome (ADAS-Cog) was not validated by the co-primary global measure Clinical Dementia Rating Scale–sum of boxes.

Differential dropout is a concern in every AD treatment trial, and could promote a positive outcome if drug treated patients drop out earlier than placebo treated patients, especially if both groups decline from the start as depicted by Dr. Crane. The fact that both the donepezil and placebo groups improved over baseline mitigates the concern that degeneration in the placebo group alone could have yielded a positive treatment effect on this one measure.

Since most of the other outcome measures in our trial did not support a treatment effect and we have reported this as a negative study, the use of LOCF analysis is not a concern for our trial. This study adds to the existing evidence that MCI should not be routinely treated with anti-dementia drugs and raises the hypothesis that different outcome measures, as well as different study designs and methods of analysis, may be necessary for future MCI treatment trials.

References

2. Salloway S, Ferris S, Kluger A et al. Efficacy of donepezil in mild cognitive impairment Neurology 2004;63:651-657.

3. Petersen R, Vitamin E and donepezil for the treatment of mild cognitive impairment: Alzheimer's Disease Cooperative Study Group. NEJM 2005:352:2379-2388.