Gabapentin and Risk of Pancreatic Cancer and Renal Cancer (GPRD): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Gabapentin and Risk of Pancreatic Cancer and Renal Cancer (GPRD) (NCT01138124), a registered phase not stated study. The primary outcome is Number of Pancreatic Cancer Cases and Matched Controls With the Indicated Exposure to Gabapentin.

2. Clinical Question

The study's stated aim was to assess whether gabapentin exposure was associated with increased risk of developing pancreatic cancer or renal cancer in the UK GPRD. The brief summary gives context for examining these outcomes: it describes animal tumor findings at high doses and a prior cancer-screening study that raised a renal cancer association, while noting possible confounding in that earlier association. Those statements explain the motivation; they are not results from this study.

The data source was a general practice database that the registry describes as centralizing information from general practitioners, specialist referrals, and hospital attendances. The target question is therefore about associations observable in this database and does not, by itself, establish that gabapentin exposure caused a cancer outcome. The specific numerical association estimates from the study are not reported in the registry record.

3. Trial Design

This is an observational case-control analysis nested in a GPRD cohort, not a randomized comparison. The record describes a source cohort spanning 1993-2008. People with incident pancreatic cancer or renal cancer were treated as cases, and matched controls were selected from the cohort. For pancreatic cancer, the case index date was the date of incident diagnosis; a control's index date was set so that follow-up from cohort entry matched the case's follow-up. The detailed matching criteria beyond that timing are not reported in the registry record.

The registry reports enrollment of 54202. That figure does not establish how many people contributed to each cancer analysis, how many cases were identified, or how many matched controls were included. Those analysis-specific counts are not reported in the registry record. The brief description says follow-up began at cohort entry and ended on Dec 31, 2008, or earlier in specified circumstances, but its population description is incomplete. Further eligibility rules and handling of people with incomplete records are not reported.

The listed intervention is gabapentin prescriptions, but exposure was observed rather than assigned. The study description defines exposure without a lag as a gabapentin prescription from cohort entry to the index date. With a 2 year lag, exposure is a prescription from cohort entry to 2 years prior to the index date, intended to address prescribing for symptoms potentially related to an undiagnosed cancer. The precise prescription categories or dose thresholds used in the analyses are not reported in the registry record.

The record lists a study start of 2010-03, primary completion of 2010-08, and status COMPLETED. It does not report randomization, blinding, or a prospective treatment protocol. The specific statistical model and analysis population definitions are also not reported.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: Actual number of patients may be less, as it is possible for a patient to be represented in more than one of the four arms (See "Participant Flow: Overall Study" Table) because of the risk set sampling.

Overall Study

Analysis population for the primary outcome, as defined in the registry: Cases and controls were drawn from the GPRD study cohort. Entry into the study cohort began Jan 1, 1993, or at the time of GPRD registration if after Jan 1, 1993. Follow-up ended Dec 31, 2008, or earlier if the respective cancer was diagnosed or if the participant left the GPRD for any reason including death.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Number of Pancreatic Cancer Cases and Matched Controls With the Indicated Exposure to Gabapentin, measured over The case index date (ID) was the date of incident pancreatic cancer diagnosis ascertained in the GPRD study cohort 1995-2008. The matched control ID was the date at which the follow-up time from his/her cohort entry was the same as that for the case.

The primary outcome measure is titled as a count of pancreatic cancer cases and matched controls with indicated gabapentin exposure. It measures exposure classification within the case-control comparison: whether prescriptions were recorded during the defined observation window for cases and their matched controls. It is not itself a direct measure of a treatment effect, nor is it a count of cancer diagnoses caused by gabapentin.

The index date anchors both the cancer case and the control's exposure window. The no-lag definition includes prescriptions up to the index date. The lagged definition stops at 2 years before that date. The record says the lag was intended to control for prediagnostic prescribing for pain symptoms possibly related to cancer. This distinction matters because prescriptions close to diagnosis could reflect symptoms preceding recognition of the cancer rather than exposure that preceded the underlying disease process.

Counts of exposed cases and controls can describe how exposure is distributed across the matched groups. To estimate an association, those counts need to be interpreted in relation to the study's matching and analysis methods. The registry record does not report the numerical endpoint results, effect estimates, confidence intervals, or the detailed statistical method used. The renal cancer question is also part of the study objective, but the primary endpoint described in the record is pancreatic cancer.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

Because exposure was not assigned, the design compares people with a cancer outcome against matched controls drawn from a database cohort. In general, an observational comparison can identify an association in recorded data, but it cannot automatically separate an exposure effect from differences between exposed and unexposed people. Factors related to both prescribing and cancer can confound the association. The registry's background specifically mentions cigarette smoking, hypertension, and lifestyle as possible confounders in a prior renal cancer association. Whether these or other factors were measured or adjusted for in this study is not reported in the registry record.

What does matching contribute?

Matching makes cases and controls comparable on the characteristics used to form the matched sets. The record describes matching controls on follow-up time from cohort entry to the index date, but does not report other matching variables, the number of controls per case, or how unmatched records were handled. In general, an analysis should account for the matching structure rather than treat all participants as unrelated observations. The specific method used in this study is not reported, so no particular regression model or variance calculation can be attributed to it.

Matching does not ensure that the groups are comparable on characteristics that were not matched or otherwise controlled. Nor does it by itself remove confounding. The quality of the association estimate depends on the matching variables, exposure ascertainment, outcome ascertainment, and analysis. Information about validation procedures for cancer diagnoses or prescriptions is not reported in the registry record.

Why define exposure with a lag?

The record distinguishes prescriptions through the case index date from prescriptions ending 2 years before it. The stated rationale for the lag is to avoid counting prescribing for pain symptoms possibly related to cancer before diagnosis. In general, this is one way to examine whether exposure close to diagnosis could be influencing the observed association. A lag can also exclude exposure that is genuinely relevant, so it changes the question being estimated. It cannot establish when a cancer began, and it does not guarantee that prediagnostic effects have been removed. The record does not report how exposure duration, dose, adherence, or multiple prescriptions were handled.

For a binary outcome, an association may be expressed through a measure comparing the odds of exposure among cases and controls, or through another model-based estimate. That is a general description, not a claim about the analysis used. The study's effect measure, model, adjustment set, and variance method are not reported in the registry record. As a result, neither an effect estimate nor its uncertainty can be reconstructed from the information reported.

How should uncertainty be read?

In general, a confidence interval communicates uncertainty in an estimated association under the chosen model. Its width depends on information in the data and the analysis. A p-value addresses compatibility with a specified null model; it does not quantify the probability that the null is true, and it does not measure clinical importance. Neither confidence intervals nor p-values for the cancer outcomes are reported in the registry record. No inferential result can therefore be interpreted from the design description alone.

The record describes a separate sample-size planning illustration based on an observed effect, using a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation. The target effect is described as a stated fraction of the observed effect. For time-to-event tables the illustration gives total events using the Schoenfeld method; other illustrations give participants per arm. These are planning illustrations, not the study's own design or observed power. The record does not report the calculated planning values, and this observational matched study should not be recast as an equal-allocation randomized trial.

Over-dispersion is a general issue when variability in count data exceeds what a simple model assumes; accounting for it can affect uncertainty estimates. The primary endpoint is described as participant counts, but the registry does not report that an over-dispersion model was used or that over-dispersion was assessed. The statistical analysis plan, software, model diagnostics, and approach to multiple comparisons are not reported in the registry record.

10. Multiplicity, Missing Data and Analysis Populations

The record lists 8 outcome measures and 80 analyses. These figures indicate that the registry contains multiple outcome and analysis entries, but do not disclose their full definitions, their relationship to one another, or which tests were designated primary. The study objective names pancreatic and renal cancer, while the specified primary endpoint is the pancreatic cancer case-control exposure count. Whether formal multiplicity adjustments were made is not reported in the registry record.

In general, testing several outcomes or exposure definitions creates more opportunities for chance findings. A multiplicity strategy can control error rates or distinguish confirmatory analyses from exploratory ones. Without the analysis hierarchy, prespecified testing plan, and adjustment details, the degree to which multiplicity affects interpretation cannot be assessed. It would not be appropriate to assume that all 80 analyses were independent tests or that an adjustment was or was not applied.

Missing-data methods are not reported in the registry record. The record does not say how missing prescription histories, covariates, or other records were handled, whether people with missing information were excluded, or whether any imputation was performed. The analyzed population for each cancer outcome is also not reported. Although enrollment is listed as 54202, the number included in each analysis and the numbers excluded for any reason are unavailable in the registry record.

In general, missing information can affect both the composition of an analysis population and the precision or validity of an estimate. Its impact depends on why the information is missing and how the analysis addresses it. Without the missingness pattern and handling rules, the direction or magnitude of any resulting bias cannot be assessed. The registry record also does not report sensitivity analyses addressing missing data or alternative analytic populations.

11. What the Result Does, and Does Not, Mean

The record describes an observational study designed to investigate whether gabapentin prescriptions were associated with pancreatic or renal cancer in a UK database cohort. It specifies how pancreatic cancer cases and matched controls were indexed and defines exposure windows with and without a 2 year lag. It also reports the enrollment figure and the study's registry dates and status. These details describe the question and broad design; they do not supply the numerical results needed to determine the observed association.

The evidence described in the registry record does not establish that gabapentin caused either cancer outcome. Observational comparisons can be affected by confounding, differences in healthcare use, exposure measurement, and other sources of bias. Matching on follow-up time addresses only the feature described in the record; additional matching, adjustment, and validation details are not reported. The lagged definition addresses a stated concern about prediagnostic prescribing but cannot by itself establish causality.

Nor does the absence of numerical cancer results in the registry record mean that no cases occurred or that no association was found. It means the numerical findings and their precision are not reported in the information available for this account. The adverse-event entries list zero participants in their reported categories, but those entries do not provide the pancreatic or renal cancer case-control outcome results and should not be substituted for them.

In general, an estimate with a confidence interval can show both the direction and precision of an association, subject to the assumptions and possible biases of the design. No such cancer estimate or interval is reported. The statistically sound interpretation is therefore restricted to the design, outcome definition, and exposure windows stated in the registry record, without a claim about the presence, absence, or size of a cancer association.

12. Why This Trial Matters Statistically

This study illustrates why the analysis design matters when an exposure is observed rather than assigned. Cases and controls were drawn from a larger database cohort, and controls were matched to cases by follow-up time to the index date. That structure makes the timing of eligibility, outcome diagnosis, and prescription history central to interpreting the comparison. A simple comparison that ignored those design features could answer a different question from the one the registry describes.

The two exposure windows also raise an important statistical issue: the definition of exposure can influence which prescriptions count in the comparison. A lagged window was included to address possible prescribing for symptoms related to an undiagnosed cancer. Such a choice can help examine reverse causation, but it is not proof that reverse causation or confounding is absent. Clear reporting of how alternative windows were analyzed is important for interpreting robustness.

The record's motivation includes a prior renal cancer association that was considered susceptible to confounding. This underscores that statistical adjustment and design features need to be reported alongside estimates. The registry record does not provide the present study's estimates, intervals, adjustment methods, or multiplicity strategy. It therefore serves as an example of how an explicit research question and an identifiable exposure definition do not, on their own, permit evaluation of evidential strength.

Finally, a registered enrollment count and a completed status do not communicate the effective information available for each cancer analysis. Case and control counts, missingness, and uncertainty are needed to judge precision. Their absence from the registry record limits statistical interpretation independently of the study's stated rationale.

13. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The most direct limitation for interpreting this record is that numerical results for the primary cancer outcome are not reported in the registry record. The numbers of pancreatic cancer cases and matched controls, exposure counts, effect estimates, confidence intervals, and p-values are not reported. The corresponding numerical renal cancer results are also not reported. Consequently, the direction and precision of the study findings cannot be assessed.

The design description is incomplete for evaluating internal validity. The record does not report detailed eligibility criteria, matching variables beyond follow-up timing, the number of controls per case, covariate adjustment, or the statistical model. It also does not report how cancer diagnoses and prescriptions were validated, how exposure duration or dose was handled, or how missing data and exclusions were managed. Enrollment of 54202 cannot be assumed to be the analysis sample for either cancer outcome.

Potential confounding is particularly relevant to an observational prescribing study. The brief summary identifies cigarette smoking, hypertension, and lifestyle as possible explanations for an earlier renal cancer association, but does not state whether those factors were available or addressed in the current analyses. Differences in prescribing or healthcare use could also matter; whether these were considered is not reported. The lagged exposure definition addresses one stated concern, prediagnostic prescribing, but does not eliminate the other limitations of observational evidence.

Multiple outcomes and analyses are listed, but the registry does not report the analysis hierarchy or multiplicity handling. The extent to which results, if available, would be exploratory or confirmatory cannot be assessed. These limitations concern the information reported in the registry record and do not establish that the study itself failed to address them.

14. Related Tutorials

15. Related Statistical Calculators

16. Sources

• ClinicalTrials.gov: Gabapentin and Risk of Pancreatic Cancer and Renal Cancer (GPRD) — NCT01138124.
• Registry record last updated 2017-05-30; status last verified 2017-05.
• Results first posted 2011-01-07.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

17. Record Summary

The record describes a completed observational study of gabapentin prescriptions and pancreatic or renal cancer in the UK General Practice Research Database. The study cohort covered 1993-2008, and enrollment is listed as 54202. The stated primary endpoint counts pancreatic cancer cases and matched controls according to gabapentin exposure. Exposure is defined both through the index date and with a 2 year lag before it.

The registry lists a study start of 2010-03, primary completion of 2010-08, and status COMPLETED. It lists 8 outcome measures and 80 analyses. Numerical cancer results, detailed statistical methods, missing-data procedures, and multiplicity handling are not reported in the registry record. The design description supports interpretation as an observational association question, not a randomized treatment comparison or a causal conclusion.