Methodological shortcomings of 12 randomised controlled trials on donepezil
Author and year | Dose | Imbalance of groups at baseline with regard to | % of missing patients in endpoint analyses—cholinesterase inhibitor (placebo)* | Missing information in publication | No correction for multiple comparisons† | Calculation of mean values may bias results | Missing information on blinding in report | Other shortcomings | ||
---|---|---|---|---|---|---|---|---|---|---|
Rogers et al 19965 | 1 mg | Weight and height | CGIC—2 (0) ADAS-cog—not reported | Correct: 1 significant result | ADAS-cog | Yes | Inconsistent reports‡ | |||
3 mg | — | ADAS-cog—not reported | ||||||||
5 mg | — | CGIC—3 (0) ADAS-cog—not reported | ADAS-cog: No of patients at end point | |||||||
Rogers et al 1998a6 | 5 mg | — | ADAS-cog—1 (2) CIBIC-plus—3 (2) | — | — | ADAS-cog CIBIC-plus | Exception: CIBIC-plus raters | Different dropout rates (P<0.05): placebo 7%; 10 mg: 18%; last observation carried forward | ||
10 mg | ADAS-cog—2 (2) CIBIC-plus—4 (2) | |||||||||
Rogers et al 1998b7 | 5 mg | — | ADAS-cog—1 (6) CIBIC-plus—3 (6) | - | - | ADAS-cog CIBIC-plus | Exception: CIBIC-plus raters | Different dropout rates (P<0.05): placebo 20%; 10 mg: 32%; last observation carried forward | ||
10 mg | Age | ADAS-cog—5 (6) CIBIC-plus—5 (6) | ||||||||
Burns et al 19998 | 5 mg | — | Not reported | ADAS-cog and CIBIC-plus: No of patients at end point | Correct: 3 significant results | ADAS-cog | Yes | Last observation carried forward | ||
10 mg | ||||||||||
Greenberg et al 20009 | 5 mg | ADAS-cog—20 (20) | — | — | ADAS-cog | — | Observed cases analysis only | |||
Homma et al 200010 | 5 mg | ADAS-cog | ADAS-cog—7 (14) J-CGIC—2 (2) | Baseline characteristics of intention to treat population | — | ADAS-cog | Yes | — | ||
Feldman et al 200111 | 10 mg | — | CIBIC-plus—3 (0) | — | — | CIBIC-plus | Yes | Yes | ||
Mohs et al 200112 | 10 mg | — | Dropout rate—28 (26) Missing patients in intention to treat population—3 (4) | Some data reported only up to week 48 | — | — | Yes | Inconsistent report‡ | ||
Tariot et al 200113 | 10 mg | Weight | Not reported | Neuropsychiatric inventory - nursing home version: No at end point | — | Neuropsychiatric inventory - nursing home version | Yes | Inconsistent report‡ | ||
Winblad et al 200114 | 10 mg | Sex | Gottfries-Brane-Steen scale—3 (0) | — | — | Gottfries-Brane-Steen scale | Yes | Inconsistent report‡ | ||
AD2000 Collaborative Group 200415 | 5 mg or 10 mg | — | Not reported adequately for the Bristol activities of daily living scale | 5 mg and 10 mg not analysed separately | — | — | — | Diverse problems, no clear duration of study, double randomisation | ||
Holmes et al 200416 | 10 mg | — | — | — | — | Neuropsychiatric inventory | Yes | Side effects of withdrawal, not efficacy of drug tested; last observation carried forward |
ADAS-cog=Alzheimer's disease assessment scale—cognitive subscale.
CIBIC-plus=Clinician's interview based impression of change with caregiver input.
CGIC=Clinical global impression of change.
J-CGIC=Japanese version of the CGIC.
↵* Refers to the intention to treat population.
↵† This criterion is satisfied, when several primary end points were calculated without correction for multiplicity and the presented results after correction exceed the significance level of 5%. To adjust for multiple comparisons we used the Bonferroni method. As many trials do not report the number of attempted comparisons, the minimum number of reported independent tests concerning primary end points was used for adjustment. For example, in the study by Burns et al,8 two dose groups of donepezil and two primary outcome measures were specified: ADAS-cog and CIBIC-plus. Therefore four comparisons were assumed for Bonferroni adjustment, leading to a required level of 0.05/4=0.0125. This, in spite of an ambiguous definition of the evaluation procedure for the CIBIC-plus in the original publication, which allowed for much more methods of comparison, all of which are mentioned in the results section of the study: comparison of means and of fractions, applying various cut-points. Assuming four comparisons, one end point did not reach the adjusted level: the comparison of donepezil 5 mg with placebo on the CIBIC-plus scores ≥3 (P=0.015). The three other results remain significant after adjustment.
↵‡ Data in tables and text, statements in text and abstract or study results, and abstracts of presentations at congresses are discrepant.