Article Text

Original article
An international randomised trial of celecoxib versus celecoxib plus difluoromethylornithine in patients with familial adenomatous polyposis
  1. Patrick M Lynch1,
  2. Carol A Burke2,
  3. Robin Phillips3,
  4. Jeffrey S Morris4,
  5. Rebecca Slack4,
  6. Xuemei Wang4,
  7. Jun Liu5,
  8. Sherri Patterson6,
  9. Frank A Sinicrope7,
  10. Miguel A Rodriguez-Bigas8,
  11. Elizabeth Half9,
  12. Steffen Bulow10,
  13. Andrew Latchford3,
  14. Sue Clark3,
  15. William A Ross1,
  16. Bonnie Malone1,
  17. Hennie Hasson2,
  18. Ellen Richmond11,
  19. Ernest Hawk12
  1. 1Department of Gastroenterology, Hepatology and Nutrition, The University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA
  2. 2Department of Gastroenterology and Hepatology, Cleveland Clinic, Cleveland, Ohio, USA
  3. 3The Polyposis Registry, St. Mark's Hospital, London, UK
  4. 4Division of Quantitative Sciences, Department of Biostatistics, The University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA
  5. 5Department of Plastic Surgery, The University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA
  6. 6Department of Cancer Prevention and Pop Science, The University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA
  7. 7Division of Gastroenterology, Hepatology and Oncology, Mayo Clinic, Rochester, Minnesota, USA
  8. 8Department of Surgical Oncology, The University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA
  9. 9Gatroenterology Department, Rambam Medical Center, Haifa, Israel
  10. 10Hvidore Hospital, Denmark, UK
  11. 11Gastrointestinal and Other Cancers Research Group, Division of Cancer Prevention, The National Cancer Institute, Bethesda, Maryland, USA
  12. 12Division of OVP, Cancer Prevention and Population Sciences, The University of Texas MD Anderson Cancer Center, Houston, Texas, USA
  1. Correspondence to Dr Patrick M Lynch, Department of Gastroenterology, Hepatology and Nutrition 1466, The University of Texas M.D. Anderson Cancer Center, 1515 Holcombe Boulevard, Houston, TX 77030-4009, USA; plynch{at}


Background and aim Although Non-steroidal anti-inflammatory drugs reduce colorectal adenoma burden in familial adenomatous polyposis (FAP), the utility of combining chemopreventive agents in FAP is not known. We conducted a randomised trial of celecoxib (CXB) versus CXB+diflouromethylornithine (DFMO) to determine the synergistic effect, if any.

Methods The primary endpoint was % change in adenoma count in a defined field. Secondary endpoints were adenoma burden (weighted by adenoma diameter) and video review of entire colon/rectal segments. Adverse event (AEs) were monitored by National Cancer Institution toxicity criteria.

Results 112 subjects were randomised: 60 men and 52 women at a mean age of 38 years. For the 89 patients who had landmark-matched polyp counts available at baseline and 6 months, the mean % change in adenoma count over the 6 months of trial was −13.0% for CXB+DFMO and −1.0% for CXB (p=0.69). Mean % change in adenoma burden was −40% (CXB+DFMO) vs −27% (CXB) (p=0.13). Video-based global polyp change was −0.80 for CXB+DFMO vs −0.33 for CXB (p=0.03). Fatigue was the only significant AE, worse on the CXB arm (p=0.02).

Conclusions CXB combined with DFMO yielded moderate synergy according to a video-based global assessment. No significant difference in adenoma count, the primary endpoint, was seen between the two study arms. No evidence of DFMO-related ototoxicity was seen. There were no adverse cardiovascular outcomes in either trial arm and no significant increase in AEs in the CXB+DFMO arm of the trial. Differences in outcomes between primary and secondary endpoints may relate to sensitivity of the endpoint measures themselves.

Trial registration number number N01-CN95040.


Statistics from

Significance of this study

What is already known on this subject?

  • The cyclooxygenase-2 inhibitor, celecoxib, reduces adenoma burden in patients with familial adenomatous polyposis (FAP) by about 30%.

  • The polyamine synthesis inhibitor, difluoromethylornithine (DFMO), when combined with non-steroidal anti-inflammatory drugs (NSAIDs), reduces tumours in animal carcinogenesis models.

  • DFMO, when combined with sulindac, reduces risk of recurrent adenomas in patients with non-familial adenomas by about 70%.

What are the new findings?

  • Celecoxib combined with DFMO is well tolerated in patients with FAP, with no irreversible hearing loss related to DFMO.

  • As measured by our traditional measure, adenoma count in a defined area, the combination of celecoxib and DFMO yields only a non-significant improvement compared with celecoxib alone.

  • A secondary but arguably more valid measure of adenoma burden, based on global video assessment, does demonstrate a significant incremental benefit through the addition of DFMO to celecoxib.

How might it impact on clinical practice in the foreseeable future?

  • Results of this randomised clinical chemoprevention trial in adults with FAP leave open the question of magnitude of benefit when adding DFMO to an NSAID. Because differences in outcome were importantly dependent on the method used to measure adenoma burden, even more careful attention to said methods will be critical in the interpretation of trial results in the future. Drug approval processes that increasingly call for demonstration of ‘clinical benefit’ underscore the importance of response measures.


Familial adenomatous polyposis (FAP) is an inherited susceptibility to diffuse colorectal adenomas, and to colorectal carcinoma, occurring in close to 100% of unresected colons, and caused by a germline mutation in the adenomatous polyposis coli gene located on the long arm of chromosome 5.1 To prevent cancer development, it is recommended that patients with FAP undergo colectomy with ileorectal anastomosis (IRA) or ileoanal anastomosis at a time when polyp burden, size or degree of dysplasia are beyond the scope of safe endoscopic management.2 This usually occurs during the second or third decade of life. Patients with the attenuated FAP phenotype, often associated with mutations at the 5′ terminus (exons 1–4), have fewer polyps and can often further delay colectomy.3

An effective chemopreventive agent or combination of agents with favourable toxicity profiles would be of substantial benefit to patients with FAP. Benefitting most are those with an IRA and recurrent rectal polyps, young adults with intact colon who wish to delay colectomy, as well as patients with duodenal adenomas. If effective in FAP, such a drug or drugs may also be efficacious in the much larger population of patients with sporadic adenomas who are at increased risk of colorectal cancer.

Aspirin and other conventional non-steroidal anti-inflammatory drugs (NSAIDs) are associated with a decreased incidence of colorectal neoplasia, but the risk of known side effects of this drug class diminishes their utility as chemopreventive agents. Selective COX-2 inhibitors were developed to minimise these toxicities, particularly GI mucosal injury and its complications, a consequence of NSAID inhibition of COX-1.4 ,5 Celecoxib (CXB) preferentially inhibits cyclooxygenase-2 (COX-2) over COX-1. COX-2 is an enzyme that is rapidly induced at the site of inflammation. Elevated levels of COX-2 are found in many premalignant lesions such as colorectal adenomas.6 ,7 Experimental evidence suggests that COX-2 inhibitors inhibit angiogenesis and induce apoptosis.8 These mechanisms may play an important role in the prevention or regression of colorectal polyps. Animal models also support the concept that COX inhibitors may play an important role in cancer prevention.4 ,9 ,10

COX-2 inhibitors have been shown to be effective chemopreventive agents in short-term trials of both FAP11 ,12 and non-familial adenomas.13–15 However, significant cardiovascular adverse event signals arose from several of the longer term sporadic adenoma trials,16 ,17 leading to the removal of rofecoxib from the market and black box warnings for CXB as well as non-selective NSAIDs.

Difluoromethylornithine (DFMO) is an irreversible, enzyme-activated inhibitor of ornithine decarboxylase (ODC), the key enzyme in mammalian polyamine synthesis, which acts through covalent binding to the enzyme.18 ,19 By blocking the conversion of ornithine to putrescine, DFMO effectively inhibits the rate-limiting first step in the biosynthesis of polyamines.20 Although the precise role of polyamines is not known, levels of ODC are closely related to tumour promotion, and inhibition of ODC is associated with suppression of tumour development in laboratory animals. DFMO was found to be an effective inhibitor of carcinogen-induced colon tumours in rats and also to inhibit cell proliferation and possibly mucosal prostaglandin levels.9 ,21–23 Nonetheless, the precise pathways by which ODC inhibition translates into tumour inhibition remain unclear. In comparison to various NSAIDs, DFMO significantly inhibits the progression of azoxymethane (AOM)-induced aberrant crypts in rat, though somewhat less effectively than sulindac or piroxicam.23 DFMO has also been studied in animal models in combination with NSAIDs. Low-dose DFMO and low-dose piroxicam were found to have an additive effect in chemoprevention of AOM-induced carcinomas in rats. The effect exceeded that of high-dose piroxicam or high-dose DFMO alone.9 ,24 Efficacy of the combined regimen in inhibiting carcinogen-induced colon carcinogenesis in rats was substantiated in other studies in rodents.23 ,24 Whether the effect of combined drug treatment reflects additive or synergistic mechanisms is not yet known. Under specific conditions, ODC induction by bile acids or growth factors (epidermal growth factor) may be prostaglandin dependent. Such findings support the potential for synergistic chemopreventive effects by DFMO and NSAIDs such as CXB.

DFMO has undergone phase I and II colon studies and studies focused on other organ sites (cervix, bladder, prostate) and was found to be well tolerated at the dose of 0.5 g/m2/day for 10 months25 ,26). As a cancer chemotherapeutic agent, the major dose-limiting effects of DFMO have been thrombocytopenia, GI toxicity (nausea, epigastric pain, stomatitis and diarrhoea) neutropenia and reversible, dose-related ototoxicity.27

Experimental evidence indicates that a combination of chemopreventive agents is a very promising strategy against colorectal cancer. Studies in the Apc Min mouse indicate that combined treatment with the NSAID piroxicam and DFMO was more effective than either agent alone.28 In another study, aspirin combined with DFMO had a synergistic effect against AOM-induced colon tumours in rats.

The key study that bears on the trial reported here is that of Meyskens et al29 in patients with non-familial adenomas. Using a combination of sulindac, 150 mg/day, and DFMO, 500 mg/day, versus placebo (PBO), a dramatic (>70%) reduction in the rate of recurrent adenomas was observed, a magnitude of effect greater than that achieved with any single agent to date.

Recognising the polyp regressing action of CXB and the potential synergy with an agent acting by a different pathway (DFMO), we undertook a randomised clinical chemoprevention trial in adult patients with FAP.


The trial was conducted at the University of Texas MD Anderson Cancer, the Cleveland Clinic in Cleveland and St. Mark’s Hospital in Harrow, UK. Persons with FAP were recruited from the UT M.D. Anderson Cancer Center Hereditary Colorectal Cancer Registry, Cleveland Clinic David G Jagelman Inherited Colon Cancer Registry and the St. Mark’s Hospital Polyposis Registry, and from outside referrals. The trial was listed with the Clinical Trials Registry ( Eligible participants aged 18–65 years were required to have a clinical diagnosis of FAP based on a personal history of one of the following: (a) >100 adenomas; (b) >10 adenomas before age 40 years; (c) >25 adenomas and, if age >40 years, a characteristic family history (autosomal dominant pattern) including >100 polyps in a first-degree family member, >25 polyps in two relatives in two generations, genetic diagnosis in a relative; or (d) genetic diagnosis by sequencing or similar assay. Participants were required to have an evaluable colon and/or rectal segment, and five or more colorectal polyps ≥2 mm at baseline examination. Additional eligibility requirements included willingness to abstain from use of NSAIDs, including aspirin, for the duration of the study (cardioprotective dose of aspirin, ≥80 mg, was permitted, following review/approval by the principal investigator (PI)), and agreement to use adequate contraception.

Exclusion criteria included (a) clinically significant hearing loss (defined as affecting everyday life or use of hearing aide); (b) participants whose air conduction pure tone audiogram revealed a deficit differing from age-specific norm by >30 dB averaged across two contiguous test frequencies in either ear; (c) unwillingness or inability to sign informed consent; (d) anticipated colectomy, within 8 months of randomisation, based on heavy polyp burden at baseline; (e) history of hypersensitivity to COX-2 inhibitors, sulfonamides, NSAIDs or salicylates; (f) chronic use of NSAIDs, including aspirin or CXB, at any dose during the 6 months prior to study entry (willingness to undergo a 3-month washout period restored eligibility); (g) use of fluconazole, lithium or chronic use of adrenocorticosteroids; (h) history in the preceding year of discrete gastric or duodenal ulcer of size >5 mm (those with Helicobacter pylori-related peptic ulcer became eligible for study upon successfully completing antibiotic treatment); (i) invasive carcinoma in the past 5 years other than resected TNM T1-2, N0 colorectal cancer or resected non-melanoma skin cancer; (j) partial or complete colectomy within 12 months prior to enrolment; (k) inability to return for follow-up tests; (l) significant medical or psychiatric problems (including significant renal, hepatic or haematological dysfunction) that would make the individual a poor protocol candidate; (m) use of any investigational agent within the last 3 months or at the discretion of the medical monitor; (n) history of pelvic radiation; (o) haemoglobin <10 g/dL, platelet count <100 000/mL; white blood cell with differential <3000/mL; serum glutamate pyruvate transaminase >1.5×upper limit of normal, serum glutamic oxaloacetic transaminase >1.5 upper limit of normal, alkaline phosphatase >1.5×upper limit of normal, bilirubin >2×upper limit of normal, creatinine >1.5×upper limit of normal; (p) positive serum pregnancy test within 14 days prior to baseline randomisation; and (q) known or prior coagulopathy.

Randomisation details: This trial was designed as a double-blinded randomised trial to compare the per cent reduction in polyp counts between patients treated with CXB with either DFMO (CXB+DFMO) or PBO (CXB+PBO) after 6 months of therapy. The design called for a total of 120 patients to be randomised and divided in 1:1 ratio between the two arms. A balanced randomisation list was generated in RANLIST ( with random balance points every 4–14 patients, and this list was provided to the lead institution study pharmacist by the Department of Biostatistics from a statistician not associated with the analysis of the trial. The pharmacist kept this list in a locked cabinet. As patients were enrolled by the research nurse at the local site, the local site pharmacist would contact the lead site pharmacist for treatment assignment. The pharmacist at each institution was to maintain the confidentiality of the treatment assignment, dispense the assigned drug and be the only entity that had access to unblinded status of individual patients during the course of the trial. The protocol provided for emergency access to this information by PI in the event of serious adverse event potentially related to drug. No such emergency access was called for during the trial period.

Endoscopic procedures: Baseline on-study colonoscopy or sigmoidoscopy as well as off-study examinations were performed with standard videoendoscopes. Video of entire withdrawal was recorded. Still photos of informative reference polyps clusters were captured. When such clusters were not adjacent to an existing anatomic landmark, such as appendiceal orifice, ileocaecal valve, surgical anastomosis or retroflex view of rectum, an India ink tattoo was placed and photographed with the polyp cluster. Polyp diameters were estimated with open or closed biopsy forceps.

Polyps were measured and reported in three ways. First, and corresponding to the primary endpoint, total polyps were counted in defined anatomic fields, using still-colour photographs, before and after treatment. As a secondary endpoint, we calculated polyp burden. Polyp diameter was recorded in size categories 0–1, 1–2, 2–3, 3–4, 4–6 and >6 mm. The total polyp burden for each patient was calculated by multiplying the number of polyps in each size category by the corresponding median polyp size for that category (or actual diameter for polyps >6 mm) and then summing over all the size categories in all matched regions.

Another secondary endpoint was a video rating of global polyp burden at 6 months compared with baseline. Five independent reviewers compared paired videos while blinded to order (pretreatment vs post-treatment) as well as to treatment arm. The first video in the pair could take the value of −2 (much better), −1 (better), 0 (same), 1 (worse) or 2 (much worse) relative to the second video. Videos from each of the four regions (caecum-ascending, transverse, descending-sigmoid, rectum) for each patient were separately scored and then averaged for patients with intact colons, and only the rectal segment was scored for postcolectomy subjects.

Interventions were as follows: (1) CXB, 400 mg orally twice a day plus DFMO PBO; or (2) CXB, 400 mg orally twice a day plus DFMO 0.5 g/m2/day rounded down to the nearest 250 mg dose (bovine serum albumin (BSA) of <1.4=500 mg/day; BSA of 1.5–2.0=750 mg/day; BSA of 2.1–2.5=1000 mg/day; BSA of >2.6=1250 mg/day).

In addition to the primary efficacy endpoint, the primary safety endpoint was relative tolerability and safety of CXB+DFMO.

Patients who dropped out after study drug administration due to adverse events or other reasons were not replaced. Those exiting the trial early were encouraged to complete the end-of-study procedures.

Following reports of COX-2 inhibitor-associated cardiovascular toxicities, the trial was suspended from 17 December 2004 to 18 March 2005, pending re-evaluation of cardiovascular risks. At one site, the trial was terminated at this point as IRB approval was not gained for it to be resumed. However, the trial resumed at the other two sites, following the addition of exclusion criteria intended to address these issues: (a) elevated C-reactive protein (>3.0 mg/L); (b) history of cardiovascular diseases or risk factors that might include one of the following: myocardial infarction, angina, coronary angioplasty, congestive heart failure, stroke or coronary bypass surgery; (c) uncontrolled hypertension (>135/>85 mm Hg on three repeated measurements during the 6 weeks prior to enrolment on the study). Eligibility was considered restored following successful treatment of known hypertension. Subjects with hypertension established for the first time at study entry were otherwise evaluated for protocol, randomised if they agreed to be monitored for BP, and were allowed to remain on-study for 3 months while undergoing antihypertensive therapy and monitoring. If, at the end of 3 months, subjects could not demonstrate successful BP control as measured and documented locally, dosing was suspended. Such subjects were nevertheless urged to complete 6-month off-study evaluation for intention-to-treat analysis; (d) uncontrolled diabetes (subjects with established or new diagnosis were managed in a fashion similar to that for hypertensives, above; (e) uncontrolled hypercholesterolaemia (subjects with established or new diagnosis managed in a fashion similar to that for hypertensives and diabetics, above—following the updated National Cholesterol Education Program Adult Treatment Panel III Guidelines); (f) family history of premature coronary disease (ie, onset <55 years of age); (g) metabolic syndrome diagnosis; (h) history of deep venous thrombosis, pulmonary embolism, systemic lupus erythematosus, family history of protein S or C deficiencies, prior heparin-induced thrombocytopenia, Factor V Leiden deficiencies or high homocysteine levels; and (i) any indications for aspirin.

Statistical methods

This trial was designed to compare the per cent reduction in polyp counts between patients treated with CXB with either DFMO (CXB+DFMO) or PBO (CXB+PBO) after 6 months of therapy. A total of 120 patients were planned for enrolment in order to detect polyp count reductions of 41% from baseline to 6 months for CXB+DFMO patients compared with 28% for CXB+PBO patients. Assuming 108 of these would be evaluable (54 per group), this would provide 80% power to detect this difference assuming an SD of 24% and a two-sided 5% significance level. An interim analysis was planned when 60 (50%) patients completed the trial. Using O'Brien-Fleming stopping boundaries,30 the two arms would be declared to be different if the p value was <0.003 or 0.049 in the interim and final analyses, respectively, preserving the overall 5% significance level.

Polyp counts and polyp burden scores from baseline and 6 months (or end-of-study procedures) were matched at up to four locations in the colon (caecum-ascending, transverse, descending-sigmoid and rectum) for individuals with an intact colon or in the rectum for those with an IRA. Only fields with corresponding photos at each time point were used to calculate the per cent change. As a result, for some patients their existing polyp counts at baseline were in fields not observed at 6 months, and thus they would have had a zero polyp count at baseline for the primary analysis. To accommodate this in the per cent reduction calculations, the polyp counts were all increased by 1, at both baseline and 6 months, before calculating the per cent change for analysis. For graphing, however, the per cent change was not adjusted by adding 1 in order to maintain a meaningful visual scale.

The modified intent-to-treat (ITT measurable) population includes all patients who had polyp counts and polyp burden scores available at both baseline and 6 months (or end of study). If the 6-month or end-of-study polyp count was not available (see paragraph above), subjects were assigned a change of 0% for the primary ITT (ITT all) analysis population. The evaluable population includes all patients with complete polyp counts and who completed 80% of treatment both overall and during the final 60 days of scheduled therapy, as defined in the protocol. Patients for all three study populations were analysed according to randomised treatment arm regardless of the actual treatment received.

For video scores, we calculated the mean rating score from the five reviewers, accounting for order. The mean rating scores for each region were calculated and the overall video rating score averaged for a single score per person.

Wilcoxon rank-sum test was used to compare the counts of polyps at least 2 mm by photos since diagnostic plots indicated significant right-skewness in the data, while t tests were used to compare the polyp burden by photos and polyp burden by video between the two arms. Waterfall plots were created for each arm for the per cent reduction in polyp counts for the ITT Full population, including all polyps.

For adverse events, specific symptoms were tabulated for all toxicities due to therapy that were experienced at grade 3 or higher or were experienced by at least 5% of patients, including only the highest grade per symptom per patient. For supplemental reporting, each patient's highest grade was identified and tabulated by grade and treatment arm, regardless of symptom. Additionally, the incidence of highest toxicity grades was reported by treatment arm (as many times as the symptom occurred). These were reported overall and separately for symptoms related to study drug.

One of the known toxicities of DFMO is hearing loss, so this was carefully monitored in this trial. Eight hearing frequencies (250, 500, 1000, 2000, 3000, 4000, 6000 and 8000 Hz) were tested in each ear. The variable of interest is the change of hearing threshold in decibels (dB) between baseline and 6 months. Characterising this variable as continuous, we calculated the mean (SD), median, minimum and maximum values for each arm, and fit linear mixed models to perform a joint analysis of all frequencies.



In total, 112 patients were randomised between December 2001 and August 2008. The study was closed at this point due to slow recruitment and with enrolment target almost met, with the mutual agreement of the Data and Safety Monitoring Board, sponsor (National Institutes of Health), and recruiting centres. At one point in time, one site was found to have inadvertently left the treatment assignments unsecured, so technically this site can no longer be considered double blind. All of those who performed photo and video scoring, however, were completely blinded the entire time. For sensitivity, the data were summarised separately by institution and are presented in the online supplementary table.

Counts of patients from randomisation to evaluable status are presented in figure 1. Table 1A presents the patient characteristic information by treatment arm. There was good balance between the treatment arms for gender, race and institution. The PBO arm (CXB+PBO) had more patients who were diagnosed by the presence of >10 polyps before age 40 and more polyps at baseline overall. The CXB+DFMO arm had more patients who were diagnosed on the basis of >25 polyps after age 40 and more patients with intact colon. Several differences exist among the patients from different centres (see online supplementary table S1). For example, site 2 only enrolled patients with colectomies and tended to enrol older patients. Site 3 only enrolled white patients. Site 1 is the only institution that enrolled patients who met the FAP definition due to >25 polyps over age 40 with positive family history and also enrolled patients with more polyps at baseline.

Figure 1

Numbers of patients from randomisation to evaluable. ITT measurable means the patient had photographed polyp counts at specified landmark at baseline and 6 months. Evaluable means that the patient was ITT measurable and took 80% of the study medication both overall and in the final 60 days of the trial. CXB, celecoxib; DFMO, difluoromethylornithine; ITT, intent to treat; PBO, placebo.

Table 1

Patient characteristics by treatment arm

Polyp change

A total of 89 patients had complete polyp information. Table 2 summarises the polyp regression results by treatment group using three outcomes: (1) per cent change in polyps of at least 2 mm; (2) per cent change in total polyp burden, considering both size and number of polyps; and (3) change in polyp burden based on videos. The first two measures are quantitative measures based on the still photos, while the third measure is qualitative, but based on entire colonoscopy video that more comprehensively surveys the colon. These outcomes and the various analysis groups (ITT measurable, ITT all and evaluable) are defined in the ‘Methods’ section.

Table 2

Response by treatment arm

The CXB+DFMO group had a mean reduction of 13% (SE=10%) in the number of polyps at least 2 mm, while the CXB+PBO group had a mean reduction of 1% (SE=14%) (p=0.69). This difference was similar for evaluable patients or when missing polyp information was replaced with 0% differences (the full ITT all analysis). Total polyp burden was decreased by an average of 40% (SE 6%) from baseline to 6 months in the CXB+DFMO group and decreased by 27% (SE 6%) in the CXB+PBO group (p=0.13). This difference remained similar in the ITT all and evaluable analyses. In the video-based analysis, patients demonstrated greater improvement (−0.80 (SE 0.14)) in the CXB-DFMO arm compared with CXB+PBO (−0.33 (SE 0.15); p=0.03). This difference was attenuated in the ITT all analysis, but retained in the evaluable analysis (p=0.01). Figure 2 presents the per cent reduction in all polyp counts for each patient ordered from largest increase to greatest reduction. The red line shows a reduction of 28%, the hypothesised average reduction for the CXB+PBO arm. We saw slightly more patients with elimination of all polyps in the CXB+DFMO group and more patients with increasing polyp counts in the CXB+PBO group. Online supplementary tables present these results separately by institution (see online supplementary table S2A). The potential unblinding at site 2 was investigated, and there was no evidence that the unlocked information left the pharmacy of that institution. Online supplementary tables show that all measures have the same trends in all centres, but with a larger magnitude of difference between arms in site 2, primarily with patients on CXB+PBO doing worse. In spite of this, the videos were assessed in a fully blinded fashion at a centralised location, so were not likely affected by the potential unblinding.

Figure 2

Per cent change in the number of polyps regardless of size from baseline to 6 months for all patients. For the graphs only, patients with observed 0% change were assigned 2% so their bars could be seen. Patients without measurable polyps were assigned 0% change for ITT. All analyses and were assigned −2% for this plot so their bars could be seen and discerned as different from observed 0% change. The red line indicates the hypothesised reduction in the CXB+PBO arm of 28%. CXB, celecoxib; DFMO, difluoromethylornithine; ITT, intent to treat; PBO, placebo.


Adverse event information was included for all but one patient with missing information. Of the remaining 111 patients, 10 were documented to experience no toxicities on the trial. The counts of toxicities related to therapy are presented in table 3 for any symptoms that occurred in at least 5% of patients or for which at least one patient experienced a grade 3 toxicity. The most common symptom was mucositis/stomatitis affecting 15 and 11 patients in CXB+DFMO versus CXB+PBO, respectively. All grade 3 events related to study drug occurred on the CXB+DFMO arm and included one each diarrhoea, gout and high-frequency hearing loss. The only symptom that differed significantly between the two arms was fatigue, which was worse in the CXB+PBO arm (p=0.02). High-frequency hearing loss was of concern for patients receiving DFMO. It occurred in seven patients on the CXB+DFMO arm compared with four patients on the CXB+PBO arm (p=0.53). Other common symptoms included diarrhoea, heartburn/dyspepsia, nausea/vomiting and headache.

Table 3

Numbers of patients experiencing adverse events attributable to study drug*

Additional toxicity tables are presented in online supplementary tables S2A–S2E examining grades of symptoms and relation to study drug, disease or other, regardless of specific symptoms. Of the 10 patients with documentation of no symptoms, 8 were in the CXB+DFMO arm compared with 2 in the CXB+PBO arm (p=0.09), whereas 8 patients had grade 3 toxicities in CXB+DFMO compared with 4 patients in the CXB+PBO arm (p=0.36). A total of 62 patients had any symptoms related to study drug, 30 vs 32 for CXB+DFMO vs CXB+PBO, respectively.

Among the total 206 drug-related toxicities, 92 (45%) of them occurred in the CXB+DFMO arm and 114 (55%) occurred in the CXB+PBO arm. Also, 19 (48.7%) of the 38 drug-related grade 2 toxicities and all 3 (100%) drug-related grade 3 toxicities belong to the CXB+DFMO arm. The heterogeneity in reported toxicities, their infrequency and lack of any consistent pattern of attribution makes it unlikely that any meaningful difference in adverse events exists between the two arms of the trial, so only related toxicities are discussed further.

Symptom categories are presented in more detail in online supplementary tables S3A and S3B. The most common categories with total counts for CXB+DFMO versus CXB+PBO, respectively, are GI symptoms (15 vs 22), auditory/ear (10 vs 14), constitutional symptoms (5 vs 12; p=0.04), pain (3 vs 5) and musculoskeletal/soft tissue symptoms (3 vs 1). The distributions of symptom categories are similar across treatment arms, except for constitutional symptoms, with more events among patients receiving PBO. Any trends are in the direction of more symptoms in the CXB+PBO arm, but none are significantly different. The specific symptoms for these top five categories are specified in online supplementary table S3B.

In light of concerns for possible ototoxicity, hearing thresholds and related measures were evaluated as toxicity measures in their own right. We performed the Wilcoxon rank-sum test and found no significant difference between the two arms at any frequency level in either ear. Overall, for the left ear, there were two patients (eight observations) in the CXB+DFMO arm (patient #28: at 1000, 2000, 3000, 4000, 6000 and 8000 Hz; patient #21: at 4000 and 6000 Hz) who had an increase of >25 dB in the hearing threshold between baseline and 6 months, compared with none in the CXB+PBO arm; and for the right ear, there was one patient in the CXB+DFMO arm (patient #28: at 8000 Hz) and one patient in the CXB+PBO arm (patient #62: at 250 Hz) who had an increase of >25 dB in the hearing threshold between baseline and 6 months.

In order to simultaneously assess the effects of treatment and frequency, a linear mixed-effect model was fit on the dB change between baseline and 6 months using frequency, treatment and the interaction between frequency and treatment as covariates. Separate analyses were conducted for the left and the right ears. The analytical results showed no significant interaction between treatment and frequency for both ears, and no significant main effect due to treatment (p=0.91, left ear; and p=0.27, right ear) or frequency (p=0.39, left ear; and p=0.61, right ear).


We report findings that build upon an earlier randomised, PBO-controlled National Cancer Institution-sponsored study in FAP study participants evaluating the chemopreventive efficacy and safety of CXB (100 or 400 mg twice a day, respectively) versus PBO administered continuously for 6 months. In that study, 83 participants from two institutions were randomised to one of three arms and 78 (94%) participants completed the study. The primary efficacy variable was the per cent change from baseline in the number of colorectal polyps. And, 23 of 30 participants in the high-dose arm had a reduction in their colorectal polyp burden, an average reduction of 28% overall. The findings were sufficiently encouraging as to support the study of this agent in combination with DFMO.

The results of this study are equivocal insofar as the potential benefits of combining DFMO with CXB in the treatment of FAP. The primary endpoint, adenoma count in a defined region of the colorectum, did not yield evidence of any significant synergies. However, when the secondary endpoints of adenoma burden as measured by count × diameter and global change by video assessment were considered, significant or marginally significant improvement was observed. In our original trial of CXB versus PBO, significance was seen with each of the endpoint measures.

The differences in the video review scores per site showed consistent results among centres. Although at one site the treatment assignments were inadvertently left unsecured by the pharmacist, technically rendering this site no longer double-blind, it should be emphasised that the trialists were not aware at any time of the randomisation. Furthermore, the video reviews were performed by a panel, including experts not from the recruiting centres. The order in which the videos were played (ie, baseline or end of study) was also randomised as to which was played first. Therefore, we feel that the integrity of the video review and the trial was maintained.

What might account for the differences between the two studies? For one, the overall adenoma burden at baseline was less striking in the subjects participating in this trial compared with the original trial. This may account for the lesser treatment effect seen in the CXB+PBO arm of this trial. However, it seems to be the case that DFMO does not afford a clinically dramatic effect when combined with CXB according to any of the measures employed to detect a treatment response.

Notwithstanding the modest nature of any possible incremental benefit from the addition of DFMO to CXB, it may nevertheless be appropriate to take a hard look at the way in which treatment response is measured in FAP. While we were disappointed that our primary endpoint, polyp burden in a small, defined region, failed to reach statistical significance, we have come to regard this as a suboptimal endpoint for several reasons. Treatment response by an ostensibly quantitative measure of adenoma count in a defined area suffers from several deficiencies. It is critically dependent on comparable technique being employed in capturing an identical field of polyps. Modest variation in technique from exam to exam is difficult to eliminate in GI endoscopy. Although such effects are randomly distributed between the two arms in a blinded trial such as this, they nevertheless tend to reduce the power to detect a difference. This effect is magnified by a lower baseline polyp burden. We believe that these factors contribute to the more positive effect seen when more of the total available information is employed, as when measuring not only polyp count but also their diameter.

When a more global assessment is performed, as with our video scoring, all of the polyp data are available for consideration, even if not quantified as rigorously. Use of this method, we believe, is more clinically relevant in any event, and probably should have served as the primary endpoint in this trial. As digital video capture, video management (as with modern video editing programs) and online streaming for convenient scoring become more routinely employed, we anticipate use of entire colonoscopy findings to be the measure of response in future trials. Further, we are finding that use of a web-based program for quantifying adenoma burden can be used by multiple observers to conveniently and quickly score adenoma burden (sum of count×diameter) from colonoscopy videos in the clinical trial setting.31 As importantly, the US Food and Drug Administration (FDA) has made it clear that endpoints in the treatment of FAP must meet a measure of ‘clinical benefit’, not merely some arbitrary per cent reduction in adenoma burden. In order to satisfy such regulators as FDA, we are developing a staging system for colorectal adenoma burden in FAP in which incrementally more aggressive interventions are tied to a given stage. In this fashion, measures such as delay in time to progression will, it is hoped, serve as a more clinically relevant endpoint for treatment efficacy. At the time of this writing, a validation study of such a staging system is nearing completion. It is anticipated that this will be a helpful part of the framework for future drug trials and will otherwise provide a common language for clinicians to employ in marking progression of severity of FAP polyp burden.

Overall, differences in adverse event rates between the CXB only and CXB plus DFMO arms in this trial did not appear to be clinically or statistically significant. Because our patient population was relatively young and the trial was short in duration, we not find or expect to find any evidence of cardiovascular toxicity in either arm of the study. Midway through the trial a careful reassessment of eligibility and monitoring was necessitated following the reports of significant cardiovascular toxicities related to both rofecoxib and CXB in several well-publicised sporadic adenoma chemoprevention trials. Because of the risk–benefit balance in our special patient population, namely high risk of malignancy but lower cardiovascular risk due to short trial design and younger patients, the trial was allowed to resume following the addition of more strict exclusion criteria and monitoring measures.

Several limitations were evident in this trial. Because the comparison arm, CXB alone, has shown significant activity in FAP, we set the bar rather high insofar as expecting a significant further reduction in adenoma burden by adding DFMO. But in designing the trial we did not feel that use of a PBO control was ethical, and to do so would have still begged the question of whether addition of DFMO really afforded any benefit over and above that seen with CXB alone. Because recruitment to FAP trials can be challenging, we did include a number of subjects with a rather minimal polyp burden. The difficulty in measuring response in such patients could have been a source of type II error. Selection of a primary endpoint, polyp counts in a defined but limited field, may have been a poor choice, given alternatives of endpoints that actually used more of the information available in a given endoscopic assessment. Might the wrong agent have been paired with DFMO? In the Meyskens trial of sporadic adenomas, a significant reduction in adenoma recurrence was seen with the combination of sulindac and DFMO. Although that trial employed no sulindac comparison arm, the per cent reduction in adenoma recurrence was greater than that observed with any other single agent to date. No head-to-head trial has been conducted or is likely between sulindac and CXB. Historically, despite some differences in trial design, the per cent reduction in adenomas appears to have been greater with sulindac. Consequently, it may be that combining sulindac with DFMO could achieve a greater synergy than was seen in this trial. Such a trial combining sulindac and DFMO in FAP is actively enrolling (Al Cohen and Eugene Gerner, personal communication).


What have we learned in this trial? The combination of CXB and DFMO, in the doses employed, is well tolerated by this mainly young patient population. Hearing loss does not seem to be an issue clinically or audiometrically. The efficacy of this combination appears marginal to substantial, depending on how the effect is measured. In the future, careful attention to the means used to measure response may be as or more important than the agent(s) employed. This will be especially the case if a measure of ‘clinical benefit’ is insisted upon, as seems evident from FDA pronouncements. Despite their obvious limitations, preclinical animal trials will be a mainstay in the process of developing drug combinations for the treatment of FAP, given the extraordinary labours involved in conducting human trials of this kind.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Contributors As lead author, PML was involved in the trial design, PI for contract with sponsor, PI for performance site protocol, performed nearly all study endoscopies on patients, virtually all of whom were patients PML followed clinically. PML worked closely with the DSMB and was responsible for reports to them, and worked closely with the statisticians in interpreting the data. PML wrote the first draft of the manuscript and oversaw the subsequent editing process. CAB completed data acquisition, interpretation of data and critical revision of article. RP contributed with the planning, conduct and reporting. He also took part in the approval of the final submission. JSM wrote the statistical analysis plan, analysed the data, drafted/revised the paper and prepared DSMB reports. RS was involved in the data analysis, manuscript writing and critical review. XW contributed with the analysis and interpretation of the data, and drafting of the manuscript. JL's role was to perform the statistical data analysis and interpret data. SP contributed to the design and conduct of the study. FAS was involved in the study conception design, interpretation of data and review/approval of the manuscript. MAR-B was involved in the interpretation and manuscript review. EH contributed with data collection and interpretation, and paper development. SB participated by being one of the reviewers and by acquisition of data, and furthermore by comments to the final manuscript. AL contributed to the data acquisition, endoscopic data and manuscript review. SC and WAR contributed with the data collection, analysis and the manuscript writing. BM was involved with the data acquisition. HH worked with Dr. Burke on this study as the research nurse coordinator. She helped with the acquisition of data, acted as a reviewer and interpretation of data. ER contributed to the planning and design of this work, review and revision of the manuscript and provided input for the methods section through an accounting of the study journey for clarification of the methods and blinding issues. EH was involved in study conception, study design, study implementation, data monitoring, data analysis, data interpretation, review and revision of final manuscript.

  • Funding This trail was funded by Pfizer under the assigned award number P30CA016672, the Biostatistics Resource Group and US NCI (N01-CN95040).

  • Competing interests None.

  • Ethics approval Institutional Review Board (IRB) ethics panels approved the study protocol at each of the participating institutions.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.