Errors in analysis and reporting of time-to-event study analysis

Endometrial cancer (EC) incidences are rising globally, primarily due to aging and the growing prevalence of obesity.[1] India is a vast country with rich cultural and regional diversity, exhibiting varying patterns of gynecological cancers across different regions. According to the National Cancer Registry Program (NCRP) report for 2012-2016, endometrial cancer ranked third among gynecological cancers in India, with age-adjusted incidence rates ranging from 0.5 per 100,000 women in Manipur to a maximum of 8.0 per 100,000 in Hyderabad. The overall cumulative risk of EC in India is 1 in 190 eligible women.[2] Philips et al.,[3] conducted a retrospective study to evaluate the status of EC in Northwest India. This study was a good attempt to explain the clinical-pathological status and variations in EC survival according to risk stratification and stage. A region-specific clinical profile and treatment outcomes are crucial for understanding the status of EC due to regional diversity. However, there are some concerns related to the reporting and analysis of results in this study: The author reported in the statistical section that univariable and multivariable Cox regression analyses were performed to identify the prognostic factors. However, the results of these analyses were not presented in the manuscript. Second, the duration of retrospective data was 5.67 years. For accurately estimating the 5-year overall survival (OS) and disease-free survival (DFS), sufficient follow-up is needed to capture enough events and ensure adequate statistical power.[4] The definition of DFS used in the study seems inappropriate. DFS should be calculated from the date of surgery to recurrence, death, or last follow-up. DFS is the length of time after primary cancer treatment during which the patient survives without any signs of cancer. Thus, the calculation should start from the surgery date, not from the date of diagnosis. Furthermore, it is unclear whether attempts were made to obtain survival details from patients or their relatives either telephonically or by other means. The manuscript reported a 5-year overall survival of 85.6% [95% CI: 54.2 to 62.1 months]. However, the 95% confidence interval should correspond to the survival probability, not the median time. A similar issue was noted with the DFS. Additionally, the x-axis of the survival curve should be labeled at 12-month intervals to align with the reported 3-year, 4-year, and 5-year OS and DFS outcomes. High censoring reduces the accuracy and increases the risk of bias. The authors should report the number of censored subjects along with the number at risk in the survival curve for better clarity. Results from the log-rank test, such as “Stage I had a 5-year OS of 98.6%, while Stage II had a median OS of 45.7 months (P < 0.001),” are unclear. It is not specified whether the comparison refers to survival probabilities or median survival time. Additionally, the log-rank test compares survival curves over the entire follow-up period. To compare survival probabilities at a specific time (e.g., 5-years), a z-test is commonly used.[5] The Kaplan–Meier and log-rank analyses are univariable methods and are susceptible to bias. For example, all high-risk patients with EC in this study were obese, which could independently increase the risk of mortality. Thus, multivariable regression is a more appropriate method for analyzing retrospective data. An appropriate statistical approach provides more robust and accurate information, especially in observational studies, to minimize bias. All statistical methods reported in the statistical section should be verified to ensure their application in the study. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.