In this study, we aimed to determine an appropriate surveillance program after gastrectomy for GC using the hazard function and showed that the risk of GC recurrence after curative surgery varies significantly according to the TNM stage and other clinicopathological factors, with a higher and earlier peak recurrence risk as the disease stage progresses. Our findings may simply confirm the evident fact that recurrence occurs earlier and more frequently as GC progresses by using the new statistical method of the hazard function. However, our results have important implications for rational post-gastrectomy surveillance programs. To the best of our knowledge, this analysis using the hazard function provides evidence regarding the optimal intensity and duration of surveillance for curatively resected GC that has not been available to date.

Current NCCN guidelines recommend frequent follow-up with medical history taking and physical examination for the first 1–2 years in patients who have undergone surgery for GC [4]. Surveillance modalities, such as CT scans, clinical tests, and upper endoscopies, are recommended only if clinically indicated for Stage I GC. For Stage II/III GC, CT scans are recommended every 6 months for the first 2 years and annually thereafter for up to 5 years. The NICE guidelines do not recommend routine surveillance, including radiological surveillance aimed only at detecting recurrence [5]. Descriptions of post-gastrectomy surveillance in other guidelines [12] are somewhat ambiguous. These recommendations in various guidelines are based on retrospectively analyzed literature and not on high-level evidence from a proper analysis of the risk of recurrence.

In our study, the HR for patients with Stage I GC remained consistently low throughout the surveillance period, suggesting that intensive surveillance may be unnecessary for patients with Stage I GC, starting from the first year postoperatively, which is consistent with the NCCN guidelines [4]. Furthermore, while the overall recurrence risk for patients with Stage II GC was lower than that for patients with Stage III GC, the peak HR was three times higher than that for Stage I. Subgroup analysis showed that, regardless of adjuvant chemotherapy or D2 lymph node dissection, the recurrence risk halved beyond 2-years postoperatively, suggesting that intensive surveillance is needed for the first 2 years after surgery for Stage II GC. In contrast, for patients with Stage III GC, the HR for recurrence halved after 3 years, suggesting the need for continued intensive surveillance for at least 3-years postoperatively.

High local recurrence suppression rates with D2 lymph node dissection were reported in a 15-year follow-up Dutch clinical trial in 2010, and D2 lymph node dissection is the standard surgical approach for patients with curable GC [3]. In the subgroup analysis, patients who underwent D1 lymph node dissection showed a significantly higher peak recurrence risk than did those who underwent D2 lymph node dissection at any stage. In our study, while the peak recurrence in the D1 group occurred 3 months earlier in Stage II and 4 months earlier in Stage III than in the D2 group, the time until the recurrence risk halved was almost the same for all cohorts, suggesting no need to change surveillance intervals based on D1 or D2 lymph node dissection, for all stages of GC.

During the study period, adjuvant chemotherapy with S-1, an oral fluoropyrimidine compound, was the standard treatment for Stage II/III GC in Japan [2]. In the Stage III cohort, the peak recurrence in patients who underwent adjuvant chemotherapy occurred 10 months later than in those who did not. However, despite the different timings of recurrence peaks, HR plots for both groups almost overlapped at approximately 20 months postoperatively. Thirty months postoperatively, the HRs for recurrence in both groups were less than half of the peak value. Therefore, intensive surveillance for 3 years postoperatively seems sufficient, regardless of adjuvant chemotherapy use. However, the resurgence of the recurrence risk at 5 years postoperatively in patients with Stage III GC who did not undergo adjuvant chemotherapy cannot be ignored. For this cohort, some forms of surveillance beyond 5 years should be considered.　Considering the recurrence pattern (Fig. 4c), the result that less symptomatic distant recurrences occur earlier than locoregional recurrences confirms the importance of intensive follow-up in the early postoperative period with imaging study.

This study used the hazard function of RFS to evaluate the trend of risk of recurrence, The effects of deaths from other diseases must be considered, and caution must be exercised when interpreting the results especially in the latter period of follow-up times. As shown in Fig. 1b, the HRs after 45 months rise again. This reflects the proportion of deaths from other diseases increases compared with the number of recurrence events (Supplementary Fig. 2). Moreover, given the univariate aspect of this analysis, some biases due to each background factor must be considered when comparing between groups. In the subgroup analysis comparing D1 and D2 in the Stage I cohort (Fig. 3a), we observed a trend of increased HRs in the D1 group after 40 months postoperatively. This trend may be an effect of death due to other diseases, and the increase of them in Stage I with the D1 cohort (Supplementary Fig. 3) may be attributed to the background factor that the patients with D1 lymph node dissection were significantly older (p < 0.01) and those who may have more comorbidities. In a comparison of operation type (total or subtotal gastrectomy) in Stage I, the HRs for total gastrectomy were twice as high as the HRs for subtotal gastrectomy during the entire period. This result is also caused by a high number of deaths from other diseases in patients who underwent total gastrectomy [Total: 11.4% (46/402), Subtotal: 4.9% (60/1230)]. Although background factor bias between groups should be kept in mind, the effect of other disease deaths is minor in the early postoperative observation period. It does not sway the conclusion of the optimal intensive follow-up period of this study.

Our study has several limitations that warrant further consideration. First, the retrospective data from a single institution used here may be subject to significant bias, and the results may lack generalizability. Second, our surveillance protocol followed the JGCA guidelines, and data from this study were obtained using a relatively intensive surveillance strategy (which is also a strength of the study). Third and most importantly, our findings could not demonstrate whether postoperative surveillance truly leads to improved survival rates or patient-centered outcomes because it was not intended for that purpose. In colorectal cancer, a 2016 Cochrane analysis found no improvement in overall survival with the implementation of enhanced surveillance of post-curative surgery patients [13]. As for GC, no report has ever proven the effectiveness of intensive follow-up with evidence, and rather, there are scattered reports denying strict follow-up [14, 15]. In the current scenario, when chemotherapy has advanced dramatically with the advent of immune checkpoint inhibitors, early detection of GC relapse through intensive surveillance may lead to a more improved prognosis than ever. Prospective randomized trials comparing intensive surveillance with less intensive surveillance (e.g., testing only when symptomatic) are needed to evaluate the genuine impact of the early detection of occult recurrence on survival and quality of life.