Topical Tretinoin Therapy and All-Cause Mortality

Objective  To evaluate the relation of topical tretinoin, a commonly used retinoid cream, with all-cause mortality in the Veterans Affairs Topical Tretinoin Chemoprevention Trial (VATTC). The planned outcome of this trial was risk of keratinocyte carcinoma, and systemic administration of certain retinoid compounds has been shown to reduce risk of this cancer but has also been associated with increased mortality risk among smokers.

Design  The VATTC Trial was a blinded randomized chemoprevention trial, with 2- to 6-year follow-up. Oversight was provided by multiple independent committees.

Setting  US Department of Veterans Affairs medical centers.

Patients  A total of 1131 veterans were randomized. Their mean age was 71 years. Patients with a very high estimated short-term risk of death were excluded.

Interventions  Application of tretinoin, 0.1%, or vehicle control cream twice daily to the face and ears.

Main Outcome Measures  Death, which was not contemplated as an end point in the original study design.

Results  The intervention was terminated 6 months early because of an excessive number of deaths in the tretinoin-treated group. Post hoc analysis of this difference revealed minor imbalances in age, comorbidity, and smoking status, all of which were important predictors of death. After adjusting for these imbalances, the difference in mortality between the randomized groups remained statistically significant.

Conclusions  We observed an association of topical tretinoin therapy with death, but we do not infer a causal association that current evidence suggests is unlikely.

Trial Registration  clinicaltrials.gov Identifier: NCT00007631

The potential of retinoid compounds to prevent cutaneous malignant lesions has been of considerable interest, and some are effective for this purpose.^1-3 In an effort to clarify the role, if any, of topical retinoids in the prevention of keratinocyte carcinomas (basal cell and squamous cell carcinomas of the skin), we sought to determine whether high-dose therapy with topical tretinoin, 0.1%, could have a chemopreventive effect. In 1998, we launched the Veterans Affairs Topical Tretinoin Chemoprevention (VATTC) Trial, a randomized trial with new basal and squamous cell carcinoma as primary outcomes.

During the VATTC Trial, a closely related systemic retinoid compound was associated with an unexpected increased mortality among smokers and decreased mortality among never smokers.⁴ Increased lung cancer incidence and mortality had previously been reported with less closely related systemically administered compounds.⁵ We report the halting of the VATTC Trial intervention 6 months before its scheduled end date because mortality in the tretinoin-treated group was higher than in the vehicle control group, and our evaluation of this potentially causal association between tretinoin therapy and increased mortality.

For the VATTC Trial, participants were recruited who had at least 2 keratinocyte carcinomas not located on perianal or genital skin or in a field of previous radiotherapy during the 5 years before randomization and who were verified as being free of skin cancer at full-body skin examination by a study dermatologist at the randomization visit. Patients with predisposing genetic conditions such as xeroderma pigmentosum or basal cell nevus syndrome or who had received an organ transplant or had known arsenic exposure, mycosis fungoides, psoralen–UV-A photochemotherapy, or previous radiotherapy to the face or ears were excluded from the study, as were patients who were pregnant or nursing, unable to give informed consent, or unable to comply with study requirements as judged by the investigator at the site. Systemic chemotherapy or retinoid treatment in the last 6 months were also exclusion criteria. In addition, a 60-day waiting period without treatment was required for several topical medications before randomization. Of particular importance to this report was the exclusion of patients with a very high risk of death within 3 years, which was defined as including a history of invasive noncutaneous cancer within the last 5 years or metastatic cutaneous cancer, or other serious medical conditions such as end-stage cardiac disease, as determined by the site investigator.

Study participants were recruited at 6 VA medical centers (Durham, North Carolina; Hines, Illinois; Miami, Florida; Long Beach, California; Oklahoma City, Oklahoma; and Phoenix, Arizona) and were examined by a study dermatologist at randomization and every 6 months thereafter until the study end date. Overall leadership responsibilities were provided by the study chairman (M.A.W.) at the VA Medical Center in Providence, Rhode Island, and by the Cooperative Studies Program Coordinating Center at the VA Medical Center, Perry Point, Maryland. Random assignment was accomplished using an interactive touch-tone telephone randomization system. Clinic coordinators screened potential study participants. After determining that a patient was eligible for randomization and had provided written informed consent, the clinic coordinator called the randomization telephone at the coordinating center. A set of questions was asked to verify eligibility of the patient and, if eligibility was verified, a unique therapy number was assigned to the patient. Each therapy number was associated with 1 of the 2 treatment groups. Associations were randomly determined using permuted-block randomization with block sizes randomly determined and stratified by the participating site. The study cream (tretinoin, 0.1%, cream or vehicle control cream; OrthoNeutrogena, a Division of Ortho-McNeil Pharmaceutical, Inc, Skillman, New Jersey) was applied to the face and ears up to twice daily as tolerated, and with lesser frequency if the patient experienced bothersome dermatitis. Before distribution of the study cream to sites, the contents (active or placebo) of the donated tubes of cream was verified by the VA Cooperative Studies Pharmacy Coordinating Center for all shipments received. Assignment of uniquely numbered tubes of cream to specific patients was accomplished through the use of a telephone system linked to databases containing the contents of the tubes by tube number and treatment assignment by patient therapy number. This system ensured that each patient received active or placebo cream at each study visit appropriate to their treatment group assignment. The primary end points for the study were time to occurrence of new basal cell carcinoma and time to occurrence of new squamous cell carcinoma. Mortality monitoring was based on notification of the site coordinators by a family member of the deceased or by the coordinator's examination of a medical record (usually in anticipation of a study visit) that indicated that the patient had died. Reporting of deaths may have been delayed if the site coordinator was unable to contact the patient or the patient's family during monthly telephone calls or to schedule a study visit and, hence, categorized the patient as lost to follow-up. After the intervention was terminated, the Beneficiary Identification Records Locater Subsystem and the master death file (extracted by the VA from Social Security Administration records) were reviewed to ascertain additional deaths.

The Charlson Index of comorbidities⁶ was calculated for each patient from the VA patient treatment databases. The Charlson Index is the weighted sum of 19 indicator variables scored as 0 or 1. Each indicator variable corresponds to a medical condition. Weights vary from 1 to 6. For example, a patient with congestive heart failure (weight = 1), moderate renal disease (weight = 2), and a metastatic solid tumor (weight = 6) would have a Charlson Index score of 9. Age and smoking status were ascertained at the initial study interview. Smoking status was ascertained because of its association with risk of squamous cell carcinoma of the skin.^7,8 A substantial percentage of participants at one site (site D) may not have been systematically asked about medication use (an item in the semiannual follow-up interview); thus, analyses involving those variables excluded that site. Independent verification of data was available from site D for other variables reported herein except for self-reported smoking (an item in the baseline interview), which was only verified independently for a portion of the sample. Hence, site D was also excluded from analyses involving smoking, although these exclusions did not materially affect the results reported. Stopping rules were not established.

The Fisher exact test was used to determine P values for 2 × 2 frequency tables; the t test was used to compare group means. Cox regression models were generated using the PHREG procedure (version 8; SAS Institute, Cary, North Carolina).

Monitoring of the VATTC Trial at least annually was conducted by the Cooperative Studies Program Coordinating Center leadership and by 17 committees: the Executive Committee of the VATTC Trial, the study Data and Safety Monitoring Board, the central Cooperative Studies Program Coordinating Center Human Rights Committee (cHRC), and the institutional review boards and the research and development committees of the 6 participating medical centers and of the study chairman's medical center (VA Medical Center, Providence, Rhode Island). In addition, the Cooperative Studies Evaluation Committee approved this study at the outset and reviewed it at its midpoint to confirm that appropriate progress had been made.

Recruitment was conducted from November 1998 through November 2002, and the last randomization was in January 2003. Study participants were to use the experimental creams and to be followed up through November 15, 2004 (end-of-study date). A total of 1131 veterans were randomized in this trial, 566 to receive tretinoin and 565 to receive the vehicle control (Figure). Their mean age at randomization was 71 years; 97% were men. No unexpected adverse events were noted at any time during the trial. The cHRC received a detailed annual report on the progress of the trial and met to discuss that report annually.

The report prepared for the April 29, 2004, cHRC meeting, based on the March 15, 2004, database update, included 167 pages of tables and 4 pages of narrative, as is typical of these reports. Since the report 1 year earlier, 59 more study participants had died, for a total of 135 (Table 1). The excess deaths in the intervention group were statistically significant (82 vs 53; P = .01; χ² = 6.54). On the basis of these data, the cHRC voted to stop the study intervention but to continue follow-up in study participants through the scheduled end of the study. On the basis of the same data, the study Data and Safety Monitoring Board advocated further analysis of this finding rather than terminating the intervention. These conflicting perspectives were presented to the study sponsor (the US Department of Veterans Affairs Cooperative Study Program), which decided to terminate the intervention as recommended by the cHRC while pursuing further analysis of the unexpected finding. The participants were notified to discontinue use of the study cream during the week of May 10, 2004 (end-of-intervention date). Given the scheduled end of the study on November 15, 2004, there would be no resumption of study treatment regardless of results of further analysis.

Because mortality was not contemplated in the original study design as an end point of interest, some of the data needed to evaluate this difference were not immediately available. Efforts were made, including accessing the VA master death file, to obtain a more complete ascertainment of deaths among participants randomized into the study and the causes of those deaths. An additional 18 deaths were ascertained in the tretinoin-treated group and 17 in the control group that had not been previously identified but were discovered through searches of the Beneficiary Identification Records Locator Subsystem database, and these were included in the June 2004 Data and Safety Monitoring Board report. Subsequently, an additional 14 deaths before the end-of-intervention date were ascertained, 8 in the tretinoin group and 6 in the control group, and another 14 deaths occurred in each group between the end-of-intervention date and the end-of-study date.

Evaluation of the data for an association with total number of tubes of cream used by self-report and with the self-reported typical (modal) frequency of use are reported in Table 2. These analyses did not reveal a clear explanation for the overall observed difference in deaths between the tretinoin-treated and vehicle control groups. The strength of evidence for that difference, as measured using the P value for the comparison between randomized groups, was greatest for those in the once-daily stratum of medication use (P = .005 and .01 as of end of intervention and end of study, respectively), was weaker in the twice-daily stratum (P = .3 and .3), and was weaker still in the less-than-once-daily stratum (P = .9 and .7). If the once- and twice-daily use strata were combined, the strength of evidence as measured using the P value would be similar to that of the once-daily stratum alone (P = .006 and .008), although the risk difference would be reduced (from 11.4% to 7.5% at the end of intervention and from 11.3% to 6.2% at the end of study).

To assess the variation in deaths by study site, we examined the data to determine whether 1 or more of the 6 sites had an apparent excessive number of deaths. No such excess was detected. We noted that site C (see Table 2) had the largest difference in deaths between randomized groups but were unable to identify a specific explanation for this difference. The additional analyses performed to evaluate the distribution of the underlying mortality to determine whether overdispersion was present, that is, whether the distribution of these rates was consistent with a Poisson distribution or had a substantially larger variation, revealed no evidence for overdispersion using the GENMOD procedure (SAS Institute) (data not shown).

We examined specific causes of death in both groups (Table 3). We noted a statistically significant (P < .05) end-of-study excessive number of deaths in the tretinoin-treated group only in the vascular disorders category; however, this category accounted for only 9 of 32 excess deaths in the tretinoin-treated group, and deaths owing to myocardial infarction, arguably a vascular disorder but classified separately, were evenly balanced between groups. Other differences of more than 5 deaths between groups included non–small cell lung cancer and respiratory, thoracic, or mediastinal disorders (7 and 8 excess deaths in the tretinoin-treated group, respectively, including 3 excess deaths in the chronic obstructive pulmonary disease subcategory) and infections or infestations (6 excess deaths in the control group). Differences as of the end-of-intervention were generally similar (Table 3). Although no single cause of death could account for the overall difference in mortality, an excessive number of deaths possibly related to smoking (respiratory and vascular disorders and non–small cell lung cancer) was considered on the basis of these data.

Attention was next focused on key variables considered potentially important predictors of death. Among those were the Charlson Index, age, sex, educational achievement (high school graduate or not), and smoking status (current smoker vs not at randomization, former smoker, and pack-years of smoking). We found that the most important predictors at both the end of the intervention and the end of the study, using univariate Cox regression analysis, were the Charlson Index, age, and smoking status. The quadratic term for age was not associated with mortality and, therefore, was not kept in the final model. The number of women in this trial was small (n = 34), and the association of female sex with reduced mortality was of borderline statistical significance (P = .056). For the Charlson Index, age, and current smoking status, a small statistically insignificant imbalance was associated with worse predicted prognosis in the tretinoin-treated group (Table 4).

The association of tretinoin use with death was analyzed using multivariate Cox regression, which adjusted for the Charlson Index, age, and smoking status. The best representation of smoking status in these models was whether the participant was a smoker at randomization. In this multivariate analysis, the association of tretinoin therapy with death remained statistically significant, albeit with a confidence interval lower boundary approaching 1.0, and only slightly diminished in magnitude (Table 5). No interaction was noted between assignment to the tretinoin group and smoking in the effect on mortality.

The VATTC Trial was a 6-year randomized chemoprevention study directed at the most common malignant lesions in the United States: basal and squamous cell carcinoma of the skin. This report evaluates the association of treatment with high-dose topical tretinoin (0.1%) with mortality that was observed during this trial and resulted in halting the intervention 6 months before the end of the study.

Our evaluation is limited by the post hoc nature of these analyses; they were not planned at the outset of the study. It is possible that some of the data used for these analyses were incomplete or inaccurate. For example, we did not verify smoking status, and our information on comorbidities was derived from VA databases that may have been incomplete. Most important, mortality may have resulted from factors not assessable in this study. Strengths of this analysis include prospective collection of data, independent of and before any knowledge of the veteran's death.

Serious adverse events, death in particular, raise obvious concern when imbalanced between treatment groups in any trial. Each situation must be evaluated in the context of the appropriate criteria for considering whether the association is one of cause and effect.

We considered the possibility that topical tretinoin applied to the face and ears might be a cause of death. One study that was published during our trial involved systemic administration of isotretinoin, a closely related compound. That trial found a significant interaction between the medication and smoking in their effect on mortality, with isotretinoin therapy associated with increased mortality among smokers.⁴ Studies involving a less closely related compound, beta carotene, have also suggested an increased risk of lung cancer and risk in smokers.^5,10-12 We looked for an interaction in our data with smoking, analogous to that found in the isotretinoin trial, but could not find one. We noted that topical tretinoin has been used to treat acne vulgaris (and other indications) and in clinical trials for decades, with no suggestion of an associated mortality risk (although death would generally be rare if it occurred at all in those trials). The differences in individual causes of death were not specific. A dose-response relationship between total number of tubes of cream used and frequency of use with risk of death was not present in our data, although we could not definitively rule out a threshold effect on the basis of our data.

We also noted the published literature that described absorption of various formulations of percutaneously applied tretinoin of 1% to 8%^13-18 and that the current prescribing information of one formulation refers to absorption of up to 31% of the applied dose under certain circumstances.¹⁹ After application for at least 1 year in 1 study, absorption of less than 2% was noted.¹⁵ Several publications have noted that the amount of tretinoin absorbed does not significantly alter the endogenous levels of tretinoin and its metabolites; the concentration of absorbed tretinoin was estimated at about 1% of the endogenous concentration.^{14,17,18,20,21}

Systemically administered all-trans retinoic acid is used as therapy for acute promyelocytic leukemia and has been particularly recommended in elderly persons.²² In that context, it has been associated with retinoic acid syndrome, which is characterized by respiratory distress and fever, often in association with pulmonary infiltrates and edema, and pleural or pericardial effusions.²³ This syndrome typically develops within 10 days of treatment initiation, is treated with systemic steroids, and is associated with mortality risk of approximately 5%.²⁴ It does not seem to be a complication of maintenance systemic all-trans retinoic acid therapy. Although the information available on deaths in the study population was limited, pulmonary effects were noted among the causes responsible for excess deaths and are prominent among the manifestations of retinoic acid syndrome, but other similarities with this syndrome were not apparent.

It has been suggested that topical tretinoin might cause pulmonary disease–related death.²⁵ However, in light of the minimal effect on systemic tretinoin concentrations and the other considerations mentioned, we found it difficult to construct biologically plausible mechanisms that would explain a direct causal link between the intervention in the VATTC Trial and death owing to multiple causes, and we were unable to conceive of a plausible mechanism by which tretinoin could indirectly lead to a fatal outcome.

No early stopping rule for the mortality end point was specified in this study. The cHRC had no explicit stopping rule and in practice stopped the intervention when the P value decreased from .19 to .01. Inasmuch as they met to review data annually during the trial (a total of 5 meetings), the probability of the cHRC stopping a trial such as this because of a mortality difference at some point in the trial may have exceeded 5% as a result of the multiple examinations performed, even if there were no true causal difference between experimental groups in deaths.

Elderly populations are characterized by multiple comorbidities and substantial mortality risk. It would be unfortunate if this resulted in failure to adequately study interventions in elderly persons. Particularly for long-term studies in this population, death should be contemplated as an end point for analysis and stopping rules should be established regardless of the known safety of the intervention. This would enable appropriate α levels to be established. It is also worthwhile to prospectively collect information on key predictors of death at enrollment so that any difference in mortality between groups that may emerge can be adequately analyzed rapidly with prompt feedback to appropriate oversight committees.

We note that there were a large number of oversight committees for this trial, leading to disagreement among them, in addition to substantial administrative burden. More efficient, streamlined oversight such as by a single central institutional review board may be beneficial in multicenter studies. The challenge is to be sensitive to potential signals of harm while minimizing the disruption, burden, and potential adverse effects associated with generating false signals.

The biological implausibility, lack of specificity of causes of death, inconsistency with previous experience, weakness of other supportive evidence in our data, and weak statistical signal cast doubt on a potential causal association of topical tretinoin with death in the VATTC Trial. However, it is important to note that our population of primarily elderly men is different from those previously reported with large-scale topical tretinoin treatment and that our dosages were higher than in these previous studies. We suggest improvements in oversight of multicenter trials such as minimizing overlapping responsibilities and establishing appropriate criteria for major decisions of monitoring committees. We do not conclude that this trial provides appropriate grounds for hesitating to use topical tretinoin in clinical practice in the absence of additional evidence.

Correspondence: Martin A. Weinstock, MD, PhD, Dermatoepidemiology Unit, VA Medical Center, 111D, 830 Chalkstone Ave, Providence, RI 02908 (maw@brown.edu).

Accepted for Publication: July 18, 2008.

Author Contributions: Drs Weinstock and Bingham had full access to all of the data in the study and take full responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Weinstock, Bingham, Collins, and DiGiovanna. Acquisition of data: Hall, Eilers, Kirsner, Naylor, Kalivas, Cole, Marcolivio, and Vertrees. Analysis and interpretation of data: Weinstock, Bingham, Lew, Hall, Collins, DiGiovanna, and Vertrees. Drafting of the manuscript: Weinstock. Critical revision of the manuscript for important intellectual content: Weinstock, Bingham, Lew, Hall, Eilers, Kirsner, Naylor, Kalivas, Cole, Collins, DiGiovanna, and Vertrees. Statistical analysis: Weinstock, Bingham, and Lew. Obtained funding: Weinstock, Bingham, and Collins. Administrative, technical, and material support: Weinstock, Bingham, Hall, Kirsner, Naylor, Marcolivio, DiGiovanna, and Vertrees. Study supervision: Weinstock, Bingham, Hall, Kirsner, Naylor, Marcolivio, Collins, and Vertrees.

Financial Disclosure: Dr Weinstock has received support from Galderma Laboratories, Johnson & Johnson, and Ligand Pharmaceuticals. Dr Eilers currently owns stock in Intuitive Surgical Devices and has owned stock in Novartis Pharmaceuticals Corp in the past 5 years. Dr DiGiovanna has served as an investigator for or has received grants from Actavis US, Allergan Inc, Galderma Laboratories, and 3M Co, and has served as a consultant to or received honoraria or travel reimbursement from Allergan Inc, Roche, and 3M Co.

Funding/Support: This study was supported by the Cooperative Studies Program (CSP) of the Office of Research and Development, US Department of Veterans Affairs, and the American Cancer Society. OrthoNeutrogena, a division of Ortho-McNeil Pharmaceutical, Inc, provided the tretinoin, 0.1%, and the vehicle creams.

Role of the Sponsors: The CSP provided the services of the CSP Coordinating Center (Perry Point, Maryland) and the CSP Clinical Research Pharmacy Coordinating Center (Albuquerque, New Mexico) provided an outside review committee to initially review the study for scientific merit, monitored study progress using various review committees, and gave final approval of the manuscript.

Study Chairman's Office

Martin Weinstock (chair), Kimberly Marcolivio (national coordinator), Providence VA Medical Center, Providence, Rhode Island.

Executive Committee

Martin Weinstock, Providence; Stephen Bingham, Perry Point, Maryland; John DiGiovanna, Providence; Russell Hall, Durham, North Carolina; Mark Naylor, Oklahoma City, Oklahoma; J. Richard Taylor, Miami, Florida; Julia Vertrees, Albuquerque, New Mexico; Clifton White, Portland, Oregon.

VA Medical Centers

Durham: Russell Hall and Deborah Hannah. Hines, Illinois: David Eilers, Nadia Sakla, and Ann Kreuger. Long Beach, California: Gary Cole, Edward Jeffes, and Terri Labrador. Miami: J. Richard Taylor, Robert Kirsner, Jonette E. Kerri, Anna G. Falabella, and Margarita Givens. Oklahoma City: Mark Naylor, Mary Beth Benson, and Lisa Perry. Phoenix, Arizona: James Kalivas, Catherine Yanni, Selma Targovnik, Janet Austin, and Susan Collier.

Dermatopathologists

Michael Piepkorn, Bellevue, Washington; and Clifton White, Portland.

Data and Safety Monitoring Board

Robert Lew, Boston, Massachusetts; Irwin Braverman, New Haven, Connecticut; Bernard Cole, Lebanon, New Hampshire; Richard Kalish, Stony Brook, New York; David McLean, Vancouver, British Columbia, Canada; and Bruce Harris Thiers, Charleston, South Carolina.

Cooperative Studies Program Coordinating Center, Perry Point, Maryland

Joseph F. Collins, Stephen Bingham, Beverly Calvert, Philip Connor, Colleen Crigler, Dawn Davis, Pat Grubb, Judy Kelly, Gail Kirk, Karen Lawson, Linda Linzy, Lorrine Palmer, and Maxine Rhoads.

Cooperative Studies Program Clinical Research Pharmacy Coordinating Center, Albuquerque, New Mexico

Mike Sather, Erica Copeland, Carol Fye, William Gagne, Patricia Grimes de Naranjo, Chad Messick, Julia Vertrees.

VA Central Office

John Feussner, chief research and development officer; Timothy O’Leary, director of clinical science research and development; Grant Huang, deputy director of clinical science research and development.