|Home | About | Journals | Submit | Contact Us | Français|
Alzheimer’s disease (AD) is a progressive neurodegenerative disease leading to dementia and eventually death. The disease is named after Alöis Alzheimer (German physician) who described the characteristic plaques and tangles in the brain of a 50 year old woman dying of the condition in 1906.1 There is no known cure for the disease. The only treatments are palliative.
AD was the 8th leading cause of death in 2000; accounting for 49,558 deaths in that year.2 It threatens to exact an increasing toll as life expectancy increases. Worldwide, the number of people with AD in 2006 was estimated to be 26.6 million. That number is projected to grow fourfold by 2050 (106.8 million people).3 In the U.S. the projection is as high as 4.58 million people with AD by 2047.4 Potential savings with disease delayed or averted are large. Brookmeyer et al estimate that an intervention resulting in a 1-year delay in disease onset would produce an annual savings of about 10 billion dollars 10 years after initiation of such an intervention.
These harsh realities have led various groups to mount randomized trials aimed at assessing the utility of various drugs and regimens in preventing development of the condition. One of those efforts culminated in the Alzheimer’s Disease Anti-inflammatory Prevention Trial (ADAPT).
ADAPT was an investigator-initiated, NIH-funded, primary prevention, randomized, placebo-controlled, multicenter trial. It was designed to address the question of whether non-steroidal anti-inflammatory drugs (NSAIDs) could prevent or delay the onset of Alzheimer’s disease (AD), with the secondary objectives of determining whether the study treatments could attenuate cognitive decline associated with aging.
The drugs tested were celecoxib (Celebrex®, Pfizer) and naproxen sodium (Aleve®, Bayer).5 This paper describes design, methods, and baseline results of the trial. Particulars of the trial are given in Table 1. Table 2 gives a chronology of the trial.
Efforts to fund the trial started with a grant application submitted to the NIH in May 1997 and culminated, two submissions later, in funding (Table 3). ADAPT was an investigator-initiated trial funded via a cooperative agreement with the National Institute on Aging (NIA).
The original proposal was for a trial of ibuprofen and matching placebo. In response to suggestions from the Initial Review Group, the third submission was changed to compare ibuprofen and celecoxib to matching placebos.
The outcome measure remained constant as AD incidence. The period of time anticipated for enrollment increased from 12 months (first and second submissions) to 18 months in the funded proposal and the amount requested (5 yr totals) grew from $10,486,428 in the first submission to $25,121,319 in the funded proposal.
The approach to recruitment also changed. In the first submission recruitment was to take place in Cache County, Utah, building on work done there in another project that had characterized a cohort of nearly 5,000 elderly people.6,7 As the scope of the proposal was expanded, the approach changed to one involving four enrolling sites (one being in Cache County). Ultimately the Cache County site was dropped in favor of another site.
The original plan was to test ibuprofen and celecoxib against a matching placebo. That plan was predicated on an assumption that it would be possible to obtain the drugs and matching placebos from the manufacturers. Funding was not adequate to cover purchase of the drugs or the manufacture of matching placebo. Several months of negotiations resulted in a commitment from Monsanto/Searle (manufacturers of Celebrex, later acquired by Pharmacia, and ultimately by Pfizer) to provide celecoxib in unmarked 200 mg “bright stock” capsules, and placebo capsules that were visually indistinguishable. It became clear early on, however, that the leading manufacturer of ibuprofen was not interested in supplying their drug and a placebo. Instead, they offered to supply commercially available drug for overencapsulation, suggesting that the large “overcapsules” could be filled with inert powder to make placebo. The investigators were concerned that the large capsules would be difficult to swallow, and also that they would be fairly easy to disassemble to see if they contained a recognizable dosage of the commercial agent. A consultant suggested that 220 mg naproxen sodium tablets (Aleve®) and matching placebo had been offered for other trials by Bayer, and that the company would probably make them available for ADAPT. Expert consultants saw no reason a priori that naproxen should not be substituted for ibuprofen. The notion of preparing and packaging 3 identical appearing capsules (celecoxib, naproxen, and placebo) was dismissed as being impractical because any change in packaging or formulation would have required extended testing and expense to satisfy the FDA regarding bio-availability and packaging.
All trials represent select study populations. The requirement for informed consent alone renders them select, but they must also meet selection criteria. The eligibility and exclusion criteria for ADAPT are given in Table 6.
The approach to enrollment in ADAPT can be characterized as a "risk concentration design", i.e., one in which persons were screened to select those considered to be at increased risk for outcomes of interest.8 The efficiency of the design depends on the degree of risk concentration accomplished by the selection factors.
ADAPT attempted “risk concentration” for AD using two known risk factors, age and family history of dementia or AD. Clearly, the incidence of AD increases with age, so part of the debate had to do with where to set the lower age cutoff (there was no upper age cutoff in ADAPT). The higher the cutoff the higher the likely incidence, but the greater the difficulty in achieving the specified sample size. The grant application as submitted proposed a lower age cutoff of 72 years, but that was later lowered to 70 to increase the pool of potential recruits.
Several of the exclusions (principally the first five proscriptions in the list of exclusions in Table 6) were imposed to reduce the risk of gastro-intestinal (GI) bleeds, inasmuch as conventional NSAIDs, including naproxen, carry known risks for such bleeding. Another obvious exclusion was use of " ≥ 4 doses/wk of non-aspirin NSAIDs" since such use might well be dangerous if in conjunction with naproxen or celecoxib. The exclusion basically had the effect of excluding people with arthritis or other conditions likely to render them as “obligate” NSAID users. Because arthritis is more common in women than men13 this exclusion, no doubt, biased the available pool toward men instead of women.
Clearly, the choice of outcome measures is an important design decision in any trial. More often than not the so called primary outcome measure is also the design variable – the variable used in the sample size calculation. Investigators in ADAPT, from the outset, wanted a "clinically relevant" measure. The argument was between a measure of cognitive decline and actual diagnosis of AD. Ultimately they settled on AD even though it was obvious that the former would have produced a smaller sample size requirement. The concerns were, first, that the observational evidence base spoke much more strongly to a “protective” effect against AD dementia than to any such effect against cognitive decline. Second was a concern that the relevance of benefit in a less steep cognitive decline with age would be more difficult to translate into clinical relevance than measurable differences in AD incidence. But it was also clear that the trial would be strapped for power using AD unless it could follow large numbers of people for long periods of time – 7 years as the trial was designed. The availability of support for such a long time required an act of faith on the part of the investigators inasmuch as NIH grant funding is made available in five year intervals. A strong probability of re-funding was essential because, even if the trial succeeded in recruiting the large number of participants needed, there was little probability of demonstrating a difference with treatment – if indeed one existed at all – after the first funding cycle (2000 – 2005).
Having settled on AD as the primary outcome measure, with cognitive decline as a secondary outcome, the next problem was to develop a reasonable estimate of AD incidence in a placebo-assigned group of people aged 70 or over having a first degree relative with age-related dementia or AD. As is often the case in trials, especially in prevention trials, the investigators were forced to make an "educated guess" from published observational studies. The figures for AD incidence used were 2.5/100 for year 1; 2.75/100 for year 2; and 3.03/100 for year 3 but, as also is often the case, they proved to be overestimates.8 The observed incidence rate per 100 person years for all treatment groups combined was 0.7.14 Undoubtedly the low rate is reflective of a “healthy volunteer effect”, but it was almost certainly a consequence also of the eligibility criterion that participants have no detectable cognitive disorder at entry into the study.
The particulars for the sample size calculation underlying the trials are given in Panel F of Table 1. Relying on the assumptions stated in Table 1, the planned sample size was 2,625. Loss due to mortality was predicted at 4/100 in year 1 with an 8.5% annual increment thereafter. Probably as another reflection of the “healthy volunteer effect,” the actual observed mortality three year rate was only 1.27/100.15
The primary difficulty in a prevention trial such as ADAPT lies in identification of people at risk of the disease of interest – AD. The usual approaches to recruitment common in trials, enrolling people already diagnosed with disease, do not work in primary prevention trials. People do not usually recognize themselves as being at risk for the disease or condition of interest. They have to be identified in some way. In ADAPT, the primary means of identification was by mass mailings to people aged 70 or over. The mailings were directed to eligible beneficiaries of Medicare using addresses supplied under an agreement between the NIA and HCFA (Health Care Financing Administration), renamed CMS (Centers for Medicare and Medicaid Services).
The mailing lists comprised eligible Medicare beneficiaries who lived in areas around the field sites, as determined by zip code. All told, there were about 3.5 million mailings. This number includes 1.2 million re-mailings to people previously contacted in the initial waves of mailings. Use of the HCFA/CMS lists obligated investigators to initiate the mailing program with a letter on Government letterhead advising recipients that their names and addresses had been given to people running ADAPT, and stating explicitly that recipients were under no obligation to participate in the study (letter on ADAPT website, www.jhucct.com/adapt).
Mailings were done by a mailing service (Alliance Mass Mailing, Forest Hill, Maryland) in waves over the course of enrollment. The general approach was to start with a mailing to a few thousand people per field site and then increase the number as field sites became proficient at screening and enrollment. The first mailings occurred in late April 2001 and the last ones in early December 2004. See Figure 1 for enrollment over time.
The yield from these mailings ranged in the neighborhood of 0.8 per 1,000 mailings and around half that fraction for re-mailing to previous recipients of letters.
The vast number of people enrolled came from the HCFA/CMS mailings. Less than 5% came via other means including (depending on the field site), use of voter registration lists, local newspaper and radio ads, presentations to church groups and civic organizations, and presentations at area academic institutions, hospitals and nursing homes.
See the ADAPT website (www.jhucct.com/adapt) for prototype copies of consent forms.
Randomization was controlled by the coordinating center. Field sites requested assignment after keying essential data necessary for assignment plus data to determine eligibility and absence of excluding conditions. Randomization was in permuted blocks of sizes 7 or 14 arranged (randomly ordered) within three age strata − 70–74, 75–79, and 80+ − within field site. Participants were counted as enrolled in the treatment group to which assigned on release of the assignment to a field site.
Treatments were administered in double-masked fashion. Masking was accomplished by use of placebos matching the celecoxib and naproxen treatments. See Martin et al5 for details on the masking. People removed from treatment by study personnel during the trial remained masked to assignment. The mask for participants and study personnel remained in place until June 2007 when it was lifted via mailings to study participants revealing assignment.
Followup was independent of treatment, that is, all persons were followed via regular study visits and telephone contacts even if they were no longer adherent to the assigned study treatment. All persons were followed to a common closing date, 28 February 2007.
The data collection schedule is outlined in Table 7. Data collection was via paper forms keyed by study personnel at the field sites via dedicated PCs. Harvests of data were done monthly by coordinating center personnel.
The list of study instruments used in ADAPT is given in Table 8.
Table 9 provides a description of the study population as enrolled. Data summarized in the table were collected during the eligibility and enrollment visits (see Table 7 for data collection schedule). The median age at entry was 74. Ages at enrollment ranged from 70 – 90.
The gender mix was 54:46 male to female. The corresponding mix in the U.S. (2000 Census data) is 39:61. The under-representation of women relative to their numbers in the general population is likely the result of the exclusion of people requiring regular treatment for the pain of arthritis – a condition more common in women than men.13
The population enrolled was predominantly white (97%) in spite of efforts to recruit minorities.
Over 3/4ths of the enrolled population had some college education with over 40% having earned a college degree – roughly twice the percentages in the corresponding U.S. population for people aged 70 −74.16
Nearly 3/4ths of the population enrolled were married on entry. About 7% reported being divorced, about the same as for similarly aged people in the U.S. population (http://www.census.gov/acs/www/; accessed 31 March 2008).
About 70% of the enrolled population consumed 6 or fewer drinks per week.
Less than 3% were cigarette smokers on entry. The corresponding figures for the U.S. population for people aged 65 and over was 8.3% as per the National Health Interview Survey, January-June 2007 (http://www.cdc.gov/nchs/data/nhis/earlyrelease/200712_08.pdf; accessed 31 March 2008). Forty-four percent reported never having smoked cigarettes.
About 17% of the enrollees reported having used non-aspirin NSAID (any dose) or aspirin (>81mg/day) aspirin in the past.
The impetus for ADAPT was driven by observational data suggesting that use of NSAIDs may reduce the incidence of AD and age-related dementia.17 By the time the trial was funded, NSAIDs capable of inhibiting activity of the COX-2 enzyme – an enzyme responsible for pain and inflammation – were available18 (NDA application 20–998; FDA approval 31 December 1998; http://www.fda.gov/cder/foi/nda/98/20998AP_appltr.pdf; accessed 31 Mar 2008). As is often the case, the newest drug is the one viewed as having the most promise. The promise in this case came from the hope that this class of drugs would have less GI toxicity than non-selective COX-2 NSAIDs. That promise is what ultimately led investigators to include a member of that class in ADAPT – celecoxib. But that decision was also the source of difficulties in ADAPT. These difficulties started in September of 2002 with a petition to the Secretary of Health and Human Services (HHS) that ADAPT be stopped because of generic concerns regarding the safety and adequacy of celecoxib. That petition was ultimately rejected, but concerns regarding the safety of the COX-2 inhibiting drugs were raised anew when Merck decided to withdraw another COX-2 inhibitor, rofecoxib (Vioxx®) from the market in September of 2004 because of newly revealed cardiovascular risks. That decision prompted the need for a letter to participants in ADAPT informing them of the withdrawal and that one of the drugs in ADAPT was a member of the COX-2 inhibiting class of drugs. The ADAPT Treatment Effects Monitoring Committee (TEMC) routinely reviewed both safety and efficacy data at its regular meetings, including its meeting of 10 December 2004. On all occasions the TEMC recommended continuation of the trial without modification. The recommendation made on 10 December was based on data gathered through 1 October 2004, but it was also clear that they saw little prospect for continuing much beyond its spring meeting unless signs of benefit started to appear.
Seven days after the meeting, on the morning of Friday 17 December a Study Officer (Director of the Coordinating Center) was informed by telephone by a representative of Pfizer, the supplier of celecoxib for ADAPT, that it was going to announce later that day that a placebo-controlled trial of celecoxib for prevention of colon polyps was being stopped because excess cardiovascular risk in the celecoxib-treated group.19 That information was relayed by the director of the ADAPT coordinating center to the ADAPT steering committee later that day in a conference call. It was during that conference call that the investigators voted to suspend treatments pending further review. The option of suspending the celecoxib treatment and continuing the naproxen treatment and its placebo was considered but rejected, since doing so would have required unmasking treatment assignments. As it turned out, even if the investigators had decided to continue the treatments, they would have been ordered six days later to suspend them when the FDA issued clinical holds on all Celebrex prevention trials (www.fda.gov/bbs/topics/ANSWERS/2004/ANS01336.html; accessed 3 April 2008; investigators notified of hold by phone 23 December 2004; see ADAPT website for letter; www.jhucct.com/adapt). The hold for ADAPT was to remain in effect until the FDA was satisfied with revision of the investigator brochure and consent documents, re-consent procedures for people already enrolled, and revision of the protocol to "ensure adequate monitoring and assessment of cardiovascular events across treatment arms".
ADAPT policy was to publish results leading to any major protocol change. Hence, once investigators voted to make the suspension permanent (see Table 2 for dates), they set about preparing a paper summarizing the cardiovascular data accumulated in ADAPT. As it turned out, it would take repeated tries with various journals before results were published (see Table 10). The reasons for rejection were that the differences, though in the wrong direction, were not "statistically significant" and hence not "believable", that the results were based on raw counts without independent reading, and that the results were somehow tainted because the decision to stop was not the product of a formal recommendation from the ADAPT TEMC. Some of these issues are raised in an accompanying commentary to the ADAPT publication by Nissen and in a reply to those comments by ADAPT investigators.20,21
A meta-analysis involving cardiovascular safety data from ADAPT and five other celecoxib-placebo controlled trials was published in early 2008.26
The writing committee extends special thanks to Jane Anau of the Office of the Chair and to Alka Ahuja, Huibo Shao, and Anne Shanklin of the Coordinating Center.
ADAPT was funded by the NIH National Institute on Aging (NIA) via a cooperative agreement; grant no. U01 AG15477
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Curtis L Meinert, Professor of Epidemiology and Biostatistics, The Johns Hopkins University, Bloomberg School of Public Health, Department of Epidemiology, 615 North Wolfe Street, Baltimore Maryland 21205.
Lee D McCaffrey, Sr Research Associate, The Johns Hopkins University, Bloomberg School of Public Health, Department of Epidemiology, 615 North Wolfe Street, Baltimore Maryland 21205.
John CS Breitner, Professor and Head, Division of Geriatric Psychiatry, Department of Psychiatry and Behavioral Sciences, University of Washington School of Medicine, Director, GRECC (S-182), VA Puget Sound Health Care System, 1660 South Columbian Way, Seattle, WA 98108.