Nicotinamide for Skin Cancer Chemoprevention: The Jury Was Out and Still is.

Key Points

Introduction
Nicotinamide has attracted considerable attention as a safe and low-cost chemopreventive agent for individuals at high risk of keratinocyte cancer (non-melanoma skin cancer), following the landmark phase III (ONTRAC) trial conducted a decade ago [1]. However, a subsequent trial (ONTRANS) in immunosuppressed solid-organ transplant recipients failed to replicate these benefits, confirming existing critiques of the ONTRAC statistical analyses [2]. Its results were further limited by poor recruitment and premature termination [3]. Despite this negative study and others reporting no significant difference between nicotinamide and placebo, [4–6] enthusiasm for nicotinamide remains high. Some clinicians have incorporated nicotinamide into practice [7]—a nationwide survey of Mohs surgeons found that 76.9% recommend nicotinamide for keratinocyte cancer prevention [8].
More recently, a large retrospective study has reignited interest in this vitamin [9]. We therefore sought to critically appraise the recent study and contrast it to earlier systematic reviews comparing oral nicotinamide with placebo in high-risk populations, focusing on outcomes of reduced actinic keratoses and/or keratinocyte cancer incidence.

What Did the Recent Large Retrospective Cohort Study Show?
In this retrospective cohort study of 33,822 veterans with prior skin cancer, oral nicotinamide 500 mg twice daily for at least 30 days was compared with no nicotinamide use to assess its effect on the time to subsequent skin cancer development. Nicotinamide use was associated with a 14% overall reduction in skin cancer risk. This risk reduction was strongest for cutaneous squamous cell carcinoma with a 22% reduction. The benefit was greatest when started after the first skin cancer (~50% reduction), but diminished with later initiation. Solid organ transplant recipients showed no overall benefit. The authors concluded that nicotinamide reduces the risk of skin cancer and is most effective when introduced early.

What is the Internal Validity of this Study?
Internal validity refers to the extent to which a study’s results accurately reflect a true causal relationship, rather than being distorted by bias, confounding or other methodological errors. An early indicator of concern is that, like most observational studies in dermatology, the protocol for this study was not registered. While registering studies may seem onerous, it promotes transparency by pre-specifying the study’s hypotheses, design and primary outcomes, thereby reducing the risk of post-hoc selective reporting of outcomes, subgroups or analytic methods—key contributors to publication bias [10].
Using the appropriate reporting guideline enhances transparency, reproducibility and completeness of reporting. Although the study was reported using the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement checklist, a more suitable framework for studies based on routinely collected data is the STROBE-RECORD (Reporting of studies Conducted using Observational Routinely-collected health Data) extension [11]. This distinction is not purely procedural—the RECORD guideline was specifically developed to address key gaps in STROBE pertinent to research using routinely collected data [11], such as the study under review [9]. Of particular relevance, STROBE-RECORD requires authors to provide a complete list of codes and algorithms used to identify subjects and classify exposures, outcomes, confounders and effect modifiers. When these details cannot be reported, an explanation should be provided. Authors should also describe the data cleaning methods applied and discuss the implications of using a dataset, which was not originally designed to address the specific research question. Additionally, the discussion should consider potential sources of bias such as misclassification, unmeasured confounding, missing data and changes in eligibility criteria over time and how these factors may have influenced the study’s results.
In line with this emphasis on methodological rigor, we used the ROBINS-I version 2 (Risk Of Bias In Non-randomized Studies of Interventions) tool [12], developed by the Cochrane Bias Methods Group, to assess bias within the study. It was judged to have a serious overall risk (see eTables 1 and 2 of the Electronic Supplementary Material [ESM]) with the most prudent issues discussed below.

Confounders

Unmeasured/Residual Confounders
The authors used propensity matching on the number/timing of skin cancers along with covariates (acitretin, field therapy, chronic lymphocytic leukaemia and solid organ transplant recipient). However, important confounders that can directly influence skin cancer risk such as skin phototype, cumulative ultraviolet exposure, family history, naevus/actinic keratoses burden, socioeconomic status, dermatology surveillance intensity and degree of immunosuppression were not measured. These important determinants may have differed between groups, thereby increasing the potential for residual confounding and distorting the observed results.

Time-Varying Confounders
A time-varying confounder is a variable that changes over time, is influenced by prior exposure, and affects both future exposure and the outcome. Variables such as exposure to field therapy, acitretin dose and immunosuppressive regimes can change over time and may influence both the likelihood of nicotinamide initiation and subsequent skin cancer risk. Although the authors adjusted for these factors at baseline, they did not account for their evolving effects after baseline. More advanced statistical methods such as inverse probability of treatment weighting, the parametric g-formula or g-estimation (collectively known as g-methods) are required to properly address these time-varying confounders [13] and yield less biased estimates of the preventive effect of nicotinamide.

Confounding by Indication
In this study, nicotinamide was prescribed without a known formal indication, which makes it susceptible to confounding by indication; a form of selection bias where the reason for receiving a particular treatment influences outcome. Patients who received it may have been more health conscious and more diligent in sun avoidance and thus skew the results towards a more favorable effect of nicotinamide. Surveillance bias is also possible: patients receiving the intervention may have been followed more frequently and closely in dermatology clinics, increasing the opportunity for skin cancers to be detected compared with controls.

Other Biases

Immortal Time Bias
Cohort studies can be prone to immortal time bias where “immortal time” refers to a follow-up period where it is impossible for participants to experience an outcome. In the context of this study, the investigators defined baseline as the date of the first prescription for nicotinamide and excluded skin cancers that were diagnosed or treated within 90 days of baseline. This 90-day window represents a period of “immortal time” as patients must survive through it to be included in the study, during which they are also assumed to be cancer free. The researchers symmetrically applied this 90-day exclusion window to both the nicotinamide (exposed) and control (unexposed) groups, but symmetry alone does not necessarily eliminate this bias. The effect of excluding early events remains uncertain. If a disproportionate number of early skin cancer cases were removed from the control group, the control group’s baseline event rate would appear artificially lower, making nicotinamide seem less protective. Conversely, if more early cases were excluded from the nicotinamide group, the protective effect would be overestimated.

Selection Bias
The Veterans Affairs cohort for this study is highly homogeneous, consisting predominantly of older white men (98% male, ~95% white with a mean age of about 77 years). This limits generalisability of the findings.

Exposure Misclassification: Compliance/Adherence to Nicotinamide
A key limitation of this retrospective study was the inability to objectively measure patient adherence to nicotinamide. While a filled prescription served as a surrogate, this only confirms the patient received the medication, not that they consumed it as prescribed. This stands in contrast to the earlier prospective study, where adherence was objectively measured at over 90%. [1] Adherence to therapy is a persistent challenge in dermatology and other specialties. [14, 15] Both the World Health Organization and a recent meta-analysis estimate non-adherence to be approximately 50%—a figure unchanged over nearly two decades. [16, 17] In the absence of objective measures of adherence, it is not possible to definitively attribute any observed reduction in skin cancer incidence to nicotinamide. Moreover, as the authors acknowledge, nicotinamide is readily available over the counter, raising the possibility that patients in the control group may have accessed it outside of the Veterans Affairs records, further biasing the findings.

Outcome Misclassification
Outcomes were ascertained using International Classification of Diseases and Current Procedural Terminology billing codes rather than histopathology [9]. This approach limits diagnostic certainty, as it does not always allow differentiation between tumour types, tumour sub-types or confirmation that each case was indeed a true malignancy in the absence of tissue-based verification.

What is the External Validity of this Study?
External validity refers to the extent to which the findings of a study can be reproduced and generalised to different populations. This is a critical aspect of any study, as high-profile investigations, particularly non-randomised investigations, are often later contradicted or found to have initially overestimated effect sizes [10, 18].

Discordance with Earlier Studies: Impact of Prior Skin Cancers
The protective benefit of nicotinamide was greatest in patients with five or fewer prior skin cancers [9], contrasting with the prospective ONTRAC trial where efficacy was demonstrated in patients with a mean of eight prior skin cancers [1]. The authors attributed this divergence to a difference in measurement: the ONTRAC trial tracked the cumulative number of cancers, while this retrospective study tracked individual occurrences (the number of cancers per patient). This individual-level tracking helps prevent outliers from skewing the overall results. While this methodological approach has merit, the inconsistent results with a prospective trial must be scrutinised to ensure they are not merely a product of residual confounding or unaddressed bias.

The Role of Data Sharing and Transparency in Causal Inference
A causal inference analysis refers to statistical methods used to establish causal relationships between an exposure and an outcome. A strong causal inference strengthens external validity by providing mathematical justification that the observed results are likely to be replicable. For example, the original ONTRANS study was critiqued for limitations in its statistical rigor; when re-analysed using a Bayesian framework, the effect size was negligible, the confidence intervals were wide or the required sample size was found to be prohibitively large [2].
Propensity score matching is a causal inference technique that seeks to minimise confounding by matching patients with similar covariates—it attempts to approximate the balance achieved through randomisation. Despite its popularity, propensity score matching has several limitations. It can, paradoxically, increase imbalance—an effect known as the propensity score paradox [19]. This paradox typically occurs when extreme caliper values are applied [20], in which case alternative methods such as Mahalanobis distance matching may warrant consideration. In addition, a Bayesian analysis that integrates prior evidence [1, 3] with the current cohort data to generate an updated posterior distribution can provide valuable complementary insights [21].
Although the authors performed propensity score matching and reported absolute standardised mean differences to assess covariate balance, they did not conduct a formal evaluation of model dependence. Model dependence refers to the extent to which results are sensitive to specific mathematical models or matching methods. Robust findings should demonstrate consistency across alternative analytic approaches, indicating low model dependence.
The discussion above underscores the significant implications of analytical flexibility in observational research. Data sharing and methodological transparency are therefore essential to ensure the replicability and credibility of findings [22]. However, because the dataset consists of proprietary Veterans Affairs data that cannot be shared [9], independent investigators are unable to re-analyse the data with alternative methods, verify the findings or evaluate for model dependence.

How Does this Study Contribute to Existing Systematic Reviews and Should Guidelines be Revised?
Two systematic reviews have evaluated nicotinamide for skin cancer prevention, the most recent of which was published in 2023 [23, 24]. Both systematic reviews came to the same conclusion—the evidence for nicotinamide as a chemopreventive agent remains weak.
These systematic reviews had their limitations. In the earlier 2022 meta-analysis, the authors included 29 trials but only five trials (552 patients) contributed to the primary outcome. Importantly, three of the five trials [25–27] were not designed to evaluate skin cancers, introducing an increased risk of bias in the outcome assessment. The authors also conducted multiple subgroup analyses, increasing the chance of a spurious association. By contrast, the 2023 meta-analysis applied stricter inclusion criteria, [24] examining only four randomised controlled trials (RCTs) on nicotinamide for keratinocyte cancer prevention with confidence intervals consistent with no significant effect (Relative Risk 0.82, 95% confidence interval 0.61–1.12). [1, 3, 4, 26] Both systematic reviews were constrained by the geographical and institutional concentration of evidence—the most influential (heavily weighted) studies originated from a single Australian research group, raising concerns about generalisability. [1, 3, 4, 26]
Since the two meta-analyses were published, only two additional studies have investigated oral nicotinamide chemoprevention—the study critiqued in this article [9] and a single-arm retrospective pre-post cohort study released several months earlier [28]. The latter study was inadequately powered, lacked a control group and was rated as poor quality when assessed using the National Institutes of Health/National Heart, Lung, and Blood Institute quality assessment tool for studies with no control group (see eTable 3 of the ESM).
Would inclusion of these two studies [9, 28] alter conclusions if incorporated into a new systematic review? Given their serious risk of bias, neither of these studies would be weighted heavily enough to materially affect the cumulative evidence or overall conclusion.

What are the Directions for Future Research?
Non-randomised or observational studies are often fragile, even modest bias can decrease the validity of a study steeply [10] and cannot be mitigated simply by very large sample sizes. To rigorously establish causality between nicotinamide and skin cancer prevention, the most impactful approach is a large (adequately powered), multi-centre, double-blind RCT. Such a trial must randomly assign high-risk individuals to receive either nicotinamide supplement or a placebo. The primary outcomes, incidence and recurrence rates of keratinocyte cancer, confirmed by histopathology, must be tracked over a sustained period with objective measures of adherence. Crucially, a RCT design eliminates confounding variables and minimises selection bias by ensuring that the only systematic difference between the groups is the treatment itself, thereby providing the highest level of evidence necessary to confirm a direct causal link.

Conclusions
Whilst we commend and encourage more research into the prevention of keratinocyte cancers; bias, confounding and flexible analytical choices can generate associations that are mistakenly interpreted as causal. This is particularly problematic in observational research, where findings are prone to over-interpretation in both academic and public domains. Such fragile inferences should be validated in randomised trials or rigorously controlled experiments, where the potential for bias is minimised. Before nicotinamide can be incorporated into clinical guidelines, a formal GRADE (Grading of Recommendations Assessment, Development and Evaluation) assessment of updated systematic review evidence is needed [29]. This framework systematically appraises the quality of evidence, the relative importance of outcomes, and the balance between benefits and risks, while also considering patient values, preferences and resource implications [30]. Until better RCT evidence emerges, nicotinamide chemoprevention remains a case of the jury was out [2]—and still is.

Supplementary Information
Below is the link to the electronic supplementary material.