Results, Rhetoric, and Randomized Trials: The Case of Donepezil

Whether donepezil provides meaningful benefit to patients with Alzheimer's disease (AD) is controversial, but drug sales annually total billions of dollars. A review of data from published randomized clinical trials (RCTs) found rhetorical patterns that may encourage use of this drug.

T he proper role of cholinesterase inhibitors (ChEIs) in managing Alzheimer's disease (AD) remains debatable. One influential review states that they ''should be considered as a standard of care,'' 1 whereas another concludes that ''the scientific basis for recommendations of ChEIs for the treatment of AD is questionable.'' 2 The American Academy of Neurology Practice Parameter makes the following practice recommendation: ''Pharmacologic treatment of AD. Cholinesterase inhibitors should be considered in patients with mild to moderate AD (Standard), although studies suggest a small average degree of benefit.'' 3 To help patients and caregivers when considering ChEIs, the published randomized clinical trials (RCTs) of donepezil for patients with clinical dementia of the Alzheimer's type were reviewed and their findings summarized. In reviewing these papers, unexpected rhetorical patterns were found.
Rhetoric is an acknowledged component of biomedical writing, but routine methods of recognizing its effects are lacking. One study 4 analyzed strategies of rhetorical influence in a group of articles on headache and produced an illuminating ''inventory of discourse features'' organized on Aristotlean principles. For example, it identified the common practice of introducing an article with prevalence or cost dataF''what Aristotle called the topos of degree''Fas a logical appeal to readers, within the category of rhetorical means called invention. This critical inquiry focused on authors' efforts to persuade readers of the credibility and importance of themselves and their work. Another study 5 emphasized the potential for ''rhetorical manipulation'' by authors and called for linguistic analysis as a third arm of peer review, but its recommendation has not been widely adopted. A third study 6 documented several instances of interpretive bias that induce ''spin'' on data from the United Kingdom Prospective Diabetes Study and proposed several types, such as the ''If enough people say it, it must be true,'' the ''We've shown something here,'' and the ''What the hell can we tell the public?'' biases. Going one step farther, a fourth study 7 advises readers in its ''Users' guide to detecting misleading claims in clinical research reports'' to avoid the discussion section altogether. The goal of the current review is to present a primitive linguistic analysis of the primary research literature on donepezilFa systematic sample of prominently displayed rhetoricFalong with the RCTs' funding and findings.

METHODS
All primary analyses of placebo-controlled RCTs of donepezil for AD were identified using a PubMed search in November 2007. Trials of patients with comorbidities (e.g., Down syndrome or Parkinson's disease) and trials that did not report clinical outcomes (e.g., those that studied functional neuroimaging) were excluded. Sentences from five prominent text sites in each report were tabulated: two from the Abstract (first sentence of results and final sentence) and three from the Discussion section (first sentence and first and last sentences of the final paragraph). These sites were specified after review of the first nine publications. Recurring rhetorical elements were highlighted.
To compare results as a possible explanation for the differences in rhetoric, outcomes of all measures that were used in at least three different RCTs were tabulated. Study characteristics and sources of funding were also tabulated for each RCT.
No external funding was received for this study, nor did an institutional review board review it. The investigators were not blinded to author or sponsorship.

RESULTS
Eighteen articles [8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25] reporting data from 16 RCTs met the inclusion criteria. Three of these articles 16,18,22 were based on a single RCT. In all but one of the 18 articles, sponsorship was declared. For the remaining article, a letter was written to the author, who provided written verification of the source of sponsorship. 13 Excerpts from the selected sentences from these articles are shown in Table 1. The complete sentences are not presented, because, in some cases, it was not possible to obtain copyright permission for reproduction. Table 2 presents size, duration, donepezil doses, and funding support for each of the articles. Table 3 presents results of measures that were reported in at least three different trials, along with the scale of these measures.
Several recurrent motifs can be seen in Table 1. ''Efficacious (or effective) . . . treating . . . symptoms'' occurred four times in three early papers. [9][10][11] The phrases ''well tolerated and efficacious,'' ''well tolerated and effective,'' and ''effective and well tolerated'' appeared 11 times in five vendor-supported RCTs. 10,11,13,15,23 Drug effects were called ''significant'' 10 times, only once with mention that the significance was statistical. [8][9][10][11][20][21][22] References to confirmation of previously demonstrated efficacy were made 15 times. 9,11,12,[14][15][16][17]20,25 One publication made the strong claim that ''results such as ours raise ethical and practical concerns regarding randomization of patients with AD to placebo in clinical trials of more than a few months duration.'' 14 Another stressed the importance of continuing therapy ''long term'' on the basis of a 1-year study. 15 Seven articles' sentences endorsed the drug for unlabeled indications: one for early disease; 20 one for neuropsychiatric symptoms and in frail, older patients; 21 and five for more-severe dementia in advance of the Food and Drug Administration extension of indications to severe disease in October 2006. [16][17][18]22,23 The most recent emphasized donepezil's benefit ''throughout the course of AD'' in its final sentence, although the study reported in the article examined only severe disease. 25 All of the RCTs with these rhetorical devices were sponsored by the vendors of donepezil.
Two trials used the word ''small'' to describe the effects of donepezil, and both noted the need for better dementia treatments. 12,19 The vendors of donepezil did not sponsor either of these. A third nonsponsored paper emphasized the drug's limitations with the negative statements ''not more effective than placebo'' and ''not . . . an effective alternative.'' 24 There were no nonsponsored trials that contained the positive rhetoric identified in the sponsored trials.
The studies and their results are consistent, regardless of sponsorship. All significant differences favored donepezil except for one sponsored study, in which quality of life was significantly better on placebo or 5 mg than on 10 mg of donepezil. 9 The 70-point Alzheimer's Disease Assessment ScaleFcognitive subscale (ADAS-cog) was used in seven RCTs, and donepezil was statistically superior to placebo in all comparisons. Treatment effect, defined as the difference between donepezil and placebo at the end of the randomized comparison interval, ranged from 1.5 to 3.2 points; in the trial not sponsored by a vendor, 19 the difference was 2.17. On the 30-point Mini-Mental State Examination (MMSE), used in 13 RCTs, the treatment effect ranged from 0.68 to 1.79 points. Two of these 13 were non-vendor-supported and found statistically significant treatment effects of 0.8 and 1.55 points. 19,24 For patients with moderate dementia, the average annual change in ADAS-cog score is 7 to 11 points and in MMSE score is 2 to 4 points. 25 All treatment effects, whether vendor sponsored or not vendor sponsored, fell within this range. Two other dementia rating scales were used in sponsored and nonsponsored trials. In the Clinical Global Impression of Change, two sponsored trials found an increase of 9% and 23% in the number of subjects scoring 4 through 7 on a 7-point Likert scale, 8,13 whereas two nonsponsored trials found no treatment effect. 12,24 In the Neuropsychiatric Inventory, three of five sponsored trials showed no treatment effect, 15,17 as did two nonsponsored trials. 19,24 DISCUSSION This simple tabulation of sentences identified several rhetorical techniques, including motifs with positive messages; use of ambiguity to imply clinical as well as statistical effects; and recurring invocation of earlier evidence to establish a ''weight of rhetoric.'' These rhetorical techniques are present only in the vendor-supported trials, and they impart a clear message that these drugs produce important benefits and do so for a wide range of patients. In contrast, rhetoric in two of the RCTs not supported by the   20 A1. Improvements favoring donepezil on the Alzheimer Disease Assessment Scale-cognitive subscale were found at weeks 12  pharmaceutical industry characterize benefits as ''small,'' congruent with the American Academy of Neurology Practice Recommendation's ''small average degree of benefit,'' and only these two RCTs refer to the need for better treatments. The data from all trials, regardless of sponsorship, were highly consistent when comparison is possible (Table 3); all showed an average degree of benefit equivalent to a few months's change in the progression of AD. The major weakness of the current study is that the rhetorical analysis was not prespecified. The five text sites were selected only after several of the articles had been read. No other rhetorical techniques, such as grammatical structure or argument strategy, or any graphical techniques were prespecified or examined. These sites have not been vali-dated and may not accurately or optimally represent an article's interpretive stance. In the absence of blinding, our own interpretive biases may have influenced our choice of text sites. Nevertheless, what has been tabulated here are only the authors' own words, chosen in a systematic, context-free, reproducible way. Finally, the RCT rhetoric was not compared with the complete data in individual articles, although where comparison is possible, the findings are comparable.
A major strength of this study is its simplicity. Limiting the analysis to RCTs of a single drug for a single disease enhances comparisons of rhetoric by limiting heterogeneity of outcomes. Substantial heterogeneity remains, because ''about 23 different scales or instruments (on average 6 per  9 24 473 5, 10 Eisai Rogers (1998) 10 12 468 5, 10 Eisai Burns (1999) 11 24 818 5, 10 Eisai Greenberg (2000) 12 12 (crossover) 60 5 National Institute of Aging Homma (2000) 13 16 24 208 10 Pfizer/Eisai Tariot (2001) 17 24 290 10 Eisai/Pfizer Feldman (2003) 18 Same RCT as reference 16 Courtney (2004) 19 114 566 5, 10 National Health Service, United Kingdom Seltzer (2004) 20 24 153 10 Eisai/Pfizer Holmes (2004) 21 12 96 10 Pfizer/Eisai Feldman (2005) 22 Same RCT as reference 16 Winblad (2006) 23 24 248 10 Eisai/Pfizer Howard (2007) 24 12 272 10 Medical Research Council/Alzheimer's Association Black (2007) 25 24 343 10 Eisai/Pfizer trial) were used. . . . Most of them were not validated for the disease for which the drugs were tested and are not currently used in clinical practice'' in the RCTs of ChEIs. 26 Another strength is the reproducible analytical framework based on specified text sites in uniformly structured scientific articles, which introduces a degree of objectivity into what could easily be a hopelessly subjective inquiry. Although focus on other text sites might yield other insights, the findings here are themselves meaningful. For several reasons, vendor influence on the rhetoric of vendor-supported RCTs is a valid concern. Rhetorical enthusiasm about donepezil is present if and only if there is vendor support. Employees of the vendors of donepezil are authors of all but one of the vendor-sponsored RCTs. The rhetoric has obvious promotional value. Other promotional devices are found elsewhere in the vendor-sponsored RCTs. For example, in two early vendor-sponsored RCTs, the brand name of the drug is provided in the first line of the abstract or elsewhere on the first page. While evidently permissible in these journals, this placement has promotional value.
The recurrence across several papers of nearly identical positive messages, such as the ''efficacious . . . treating . . . symptoms'' motif, raises the possibility that company guidance may have influenced authors' choice of words and phrases or that professional medical writers may have assisted in preparation of the manuscripts. Discussing a lawsuit involving papers alleged to have been produced by drug company writers, then signed by academic physicians as their own, a medical writer in 1999 described her experience: ''I was given an outline, references, and a list of drug-company approved phrases. . . . I was pressured to rework my drafts to position the product more favorably.'' 27 The problem has been described in detail. 28 Since 2004, the World Association of Medical Editors has required that writers who ''draft or revise the article'' and their sponsors be identified. 29 There was no evidence about whether any of the RCTs discussed here were in any way ghostwritten.
Documents from a successful lawsuit against gabapentin's vendors demonstrated clearly how a vendor can develop a ''publication strategy'' with promotional intent. A medical education company's proposal to the vendor of gabapentin, for example, promised that ''all articles submitted will include a consistent message.'' It has been shown how this lawsuit uncovered ''activities traditionally considered independent of promotional intent, including continuing medical education and research, (that) were extensively used to promote gabapentin. New strategies are needed to ensure a clear separation between scientific and commercial activity.'' 30 Intentionally promotional rhetoric within the text of an RCT may be particularly effective because the difference between scientific and commercial activity is so unclear. Thus, readers of scientific articles may not anticipate embedded promotional rhetoric, because ''public discourse about this published evidence of efficacy and safety rests on the assumption that clinical trials data have been gathered and are presented in an objective and dispassionate manner.'' 31 Our hypothesis-generating study suggests that industry-sponsored authors may have taken advantage of this assumption. The uniformly favorable rhetoric in their published RCTs may have helped promote the multibillion-dollar commercial success of a drug whose clinical relevance remains uncertain.