Association between incidence of hypomagnesemia and better prognosis of NSCLC: A case on the immortal time bias

<sec>Background<p> Previous studies predominantly reported that the occurrence of hypomagnesemia was associated with improved cancer prognoses. However, utilizing publicly available data, we demonstrated the presence of bias in this association, attributable to the failure to account for immortal time bias (ITB) when grouping patients based on post-therapeutic characteristics/variates. </p></sec><sec>Methods<p> This research included 4,172 individuals from four clinical trials of Atezolizumab (NCT02366143, NCT02367794, NCT02367781, NCT02008227) in non-small cell lung cancer (NSCLC). Joint modelling, landmark, and clone-RMST (restricted mean survival time) methods were performed to remove ITB. Previous studies about hypomagnesemia and prognosis in cancers were pooled. </p></sec><sec>Results<p> Under the risk of ITB, spurious benefits in progression-free survival (PFS) and overall survival (OS) were observed in patients with at least one episode of hypomagnesemia, based on our research and previous studies. When controlling for ITB, the initially observed false beneficial effect disappeared, revealing a significant association between hypomagnesemia and worse OS [HR: 1.01 (1.00, 1.02), <i>P</i>=<styled-content style-type="number">0.0330</styled-content>] in joint modelling. A similar trend was observed in PFS [HR: 1.01 (1.00, 1.02), <i>P</i>=<styled-content style-type="number">0.0564</styled-content>]. Consistent results were obtained from landmark and clone-RMST analysis. The estimands and application scenarios of the three methods were compared and discussed. </p></sec><sec>Conclusions<p> Three methods were utilized and recommended to control ITB. Time zero definition should be emphasized to avoid ITB with advocacy of guidelines. In NSCLC, serum magnesium levels could serve as a potential prognostic biomarker in first-line therapy, but not in non-first-line therapy.</p></sec>