Population-based study of effectiveness of neoadjuvant radiotherapy on survival in US rectal cancer patients according to age

Recent cancer researches pay more attention to younger patients due to the variable treatment response among different age groups. Here we investigated the effectiveness of neoadjuvant radiation on the survival of younger and older patients in stage II/III rectal cancer. Data was obtained from Surveillance, Epidemiology, and End Results (SEER) database (n = 12801). Propensity score matching was used to balance baseline covariates according to the status of neoadjuvant radiation. Our results showed that neoadjuvant radiation had better survival benefit (Log-rank P = 3.25e-06) and improved cancer-specific 3-year (87.6%; 95% CI: 86.4–88.7% vs. 84.1%; 95% CI: 82.8–85.3%) and 5-year survival rates (78.1%; 95% CI: 76.2–80.1% vs. 77%; 95% CI: 75.3–78.8%). In older groups (>50), neoadjuvant radiation was associated with survival benefits in stage II (HR: 0.741; 95% CI: 0.599–0.916; P = 5.80e-3) and stage III (HR: 0.656; 95% CI 0.564–0.764; P = 5.26e-08). Interestingly, neoadjuvant radiation did not increase survival rate in younger patients (< = 50) both in stage II (HR: 2.014; 95% CI: 0.9032–4.490; P = 0.087) and stage III (HR: 1.168; 95% CI: 0.829–1.646; P = 0.372). Additionally, neoadjuvant radiation significantly decreased the cancer-specific mortality in older patients, but increased mortality in younger patients. Our results provided new insights on the neoadjuvant radiation in rectal cancer, especially for the younger patients.

Statistical analysis. Propensity score matching. We defined propensity score as the probability of patients being in the neoadjuvant radiation group. Propensity score model is preferable to include prognostically important covariates than these that affecting the treatment-selection process 30 . Firstly, we select these important prognostic factors: tumor size 31 , histological type 32 , differentiation 31 , age 6 , number of harvest lymph node 33,34 , tumor stage 7 . Secondly, some confounding variables which included in previous CRC 6,32,35 or propensity score based research 27 and might be related with treatment and outcome were selected. Finally, in combination with records in SEER, nine covariates: race, gender, age, tumor size, tumor number, stage, histological type, grade, lymph node resected were used to build propensity score models. The propensity scores were calculated by a nonparsimonious multivariable logistic regression model, which used neoadjuvant radiotherapy as the outcome of interest and nine clinical features as covariates. A nearest neighbor and 1 to 1 matching algorithm was performed within default caliper (0.2) in SPSS 36 . Standardized difference (S.D) between neoadjuvant radiation and no-neoadjuvant radiation patients was calculated for every baseline covariate to test corresponding balance 29 . S.D before and after matching was compared to see whether the S.D was changed to less than 0.1, which indicated a good balance for matched data 30 . A sensitivity analysis was performed to test the robustness of matched data using the R package rbounds 37 . R packages powerSurvEpi 38,39 was used to estimate the sample size necessary for 80% statistically power.
Cancer specific survival curve and Cox model construction. In propensity score matched SEER dataset, survival time was calculated from the date of diagnosis to death or to the end of study. Patients who were lost followed-up or still alive at the end of study were censored. Cancer-specific survival curve was generated using Kaplan-Meier estimate, and difference between different treatment groups were analyzed by log-rank test. Additionally, 3-year and 5-year cancer specific survival rates were calculated for neoadjuvant radiation and no-neoadjuvant radiation groups. Cox proportional hazards regression model which contained all measured covariates was performed and hazard ratio (HR) was calculated to assess the importance of neoadjuvant radiotherapy and other covariates. To assess the effects of unmeasured clinical variables on HR, sensitivity analysis [40][41][42] was performed by R package obsSens.
Scientific RepoRts | 7: 3471 | DOI:10.1038/s41598-017-02992-7 All Survival rate, HR and risk ratio (RR) were reported with their corresponding 95% CI. All statistical test were two-sided and a P value of less 0.05 was considered to be statistically significant.
Patient stratification by clinical covariates. Based on the results of Cox model, we selected clinical covariates associated with neoadjuvant radiotherapy. We grouped patients by each covariate, performed cancer specific survival analysis and constructed Cox model in each subgroup.
Effect of age on neoadjuvant treatment analysis. We divided patients into younger and older groups. For each group, annual age-adjusted incidence rates (using the 2000 U.S. standard population) were calculated of rectal cancer per 100,000 from 1992 to 2012 using SEER*Stat software version 8.3.4 43,44 . Fisher's exact test was used to compare the distribution of every clinical variable item between younger and older patients. Cancer specific survival curve and Cox proportional hazards regression were performed to evaluate the effectiveness of neoadjuvant radiotherapy. HR was calculated to assess the importance of neoadjuvant radiotherapy and also 3, 5 year survival rates were calculated in different treatment groups. The RR was calculated in different time interval for younger and older groups. We further combined age and stage to stratify patients, and investigate the effectiveness of neoadjuvant radiotherapy in four subgroups.

Results
Data source and propensity score matching. We performed a comprehensive study to access the impact of neoadjuvant radiation on survival in rectal cancer patients (Fig. 1). Firstly, clinical data of 545474 CRC patients were obtained from the SEER database. Stage-specific survivals were significantly different in five stages, which was consistent with a previous study 7 . Adults over age 50 are high-risk population of CRC who are recommended to receive CRC screening. Our data also showed that 80% rectal patients were above age 50 (see Supplementary Fig. 1AB). We also calculated annual age-adjusted incidence rate per 100,000 between younger and older groups for rectal cancer during 1992~2012. Trend of incidence rate decreased in adults over age 50, while rose for adults less 50 (see Supplementary Fig. 1C), which was consistent with the trend in previous CRC studies 19,20 . Therefore, age 50 was used to divide patients into younger (age < = 50, n = 2511) and older groups (age >50, n = 10290). Totally, There were 12801 rectal patients in stage II or III, or whom 6117 (48.25%) had neoadjuvant radiation before surgery since 2007. Their clinical characteristics were summarized in Table 1.
There were systematic differences in baseline characteristics between neoadjuvant radiation and no-neoadjuvant radiation patients in the overall samples. The S.Ds of gender, age, lymph node resected etc were larger than 0.1 between neoadjuvant radiation and no-neoadjuvant radiation patients which means the distribution of those items were unbalanced (Table 1 left). Therefore, a propensity score model was built to get a similar distribution of these measured baseline covariates. After propensity score matching, the standardized differences of all covariates were less than or close to 0.1 (Table 1 right), which indicated the covariates were balanced between two treatment groups. This balanced cohort totally had 9968 (77.8%) patients, and the number of two group patients were same (n = 4984). Furthermore, we did the sensitivity analysis to test whether the matched dataset was robust to clinical variables that were not considered in the propensity score model. The results showed our matched data was robust to the unconfouned variants (Γ = 7, P = 2.1e-05), as described in previous studies 37 .
To determine whether the sample size was enough to reach a power of 0.8, we estimated the number of patients for survival analysis and the number of deaths for Cox model. Result showed that at least 1096 patients and 302 deaths were needed in each treatment subgroup. Therefore, our propensity score matched dataset (4984 in each group) could achieve higher power.

Cancer-specific survival analysis and Sensitivity analysis. Survival benefit of neoadjuvant radia-
tion. In the propensity score matched cohort, the median follow-up time was 27 months (ranged 0-71 months). The 3-year and 5-year cancer-specific survival (CSS) rates were 87.6%(95% CI: 86.4-88.7%), 78.1%(95% CI: 76.2-80.1%) in neoadjuvant radiation group and 84.1%(95% CI: 82.8-85.3%), 77%(95% CI: 75.3-78.8%) in no-neoadjuvant radiation group. The CSS rates of neoadjuvant radiation group was significantly better than no-neoadjuvant group ( Fig. 2A, Log-rank P = 3.25e-06). The neoadjuvant radiotherapy was associated with reduced disease-specific mortality at 6 months (RR, 0.25; 95% CI, 0.18-0.35), 1 year (RR, 0.44; 95% CI, 0.35-0.55) and 2 year (RR, 0.59; 95% CI, 0.50-0.69). In order to investigate how the clinical factors jointly affected survival, multivariable proportional Cox model was constructed. When considering all clinical factors together, neoadjuvant radiation also achieved a significantly disease-specific survival benefit (HR, 0.741; 95% CI: 0.646-0.811; P = 2.27e-07). Additionally, grade, number of harvested lymph nodes, histological types, tumor size, stage and age also had significant effect on survival (Fig. 2B). Although Cox model included 23 clinical items that presumably captured most of potential cofounding, some unknown or unmeasured covariates associated with both treatment and outcome still existed. In order to eliminating the effect of unmeasured covariates, we did a sensitivity analysis for the Cox model. Supposing there existed an unmeasured covariant, we changed its parameters (unmeasured confounder HR, prevalence in neoadjuvant group, and prevalence in no radiotherapy group) to see whether we could get similar conclusion for neoadjuvant radiation (Fig. 2C). In most conditions, the unmeasured covariant did not change our conclusion (yellow). The opposite effect of neo-radiotherapy only occurred in some extreme conditions (red). For example, unmeasured confounders whose HR was 2.0 or greater and the prevalence in no radiotherapy group was larger than 0.5. Thus, the beneficial effect of radiotherapy was robust to departures from ignobility. Impact of neoadjuvant radiation on survival in different clinical subgroups. Then we grouped patients according to the clinical covariates, and compared the impact of neoadjuvant radiation on survival in different subgroups. Cancer specific survival and Cox model showed that neoadjuvant radiation played significant roles in all stage-related and age-related subgroups, but it did not significantly affect all subgroups of other clinical covariates (see Supplementary Fig. 2A1,D2, Supplementary Table S1).
In both stage II (Fig. 3A) and III (Fig. 3B) patients, the cancer specific survival of neoadjuvant radiation was significantly better than in the no-neoadjuvant group (Log rank P = 0.048, 1.29e-05). In stage II patients

Impact of neoadjuvant radiation on survival in younger and older patients. Different from
stage-subgroups, neoadjuvant radiation had opposite roles in two subgroups of age. Neoadjuvant radiation improved survival in older patients (P < 1.06e-08, see Supplementary Fig. 2E2), but no-neoadjuvant group had better survival than neoadjuvant group in younger patients (P = 0.04, see Supplementary Fig. 2E1). In order to confirm the opposite role of neoadjuvant radiation on age, we further excluded the impact of other covariates by using propensity score matching in age-related groups. Since the standardized differences of all covariates were already balanced well in older group (see Supplementary Table S2 right), there was no need to do propensity score matching. One covariate deviated from the balance status in younger group (see Supplementary Table S2 left), therefore we did propensity score matching in younger group and got a new dataset (see Supplementary Table S2 middle part). Survival analysis in this new dataset showed the same trend as the original data (Log rank younger: P = 0.022, older: P < 1.06e-08, Fig. 4A/B). In younger group (n = 1846), 3-year and 5-year survival rates were For younger group, the RR between neoadjuvant radiation and no-neoadjuvant group was less than 1 in the first two years, but it increased to more than 1 since the 3 rd -year. Neoadjuvant radiation was significantly associated  with increased disease-specific mortality in the 6 th -years. However, in older group neoadjuvant radiation was associated with reduced disease-specific mortality significantly (RR < 1) (Fig. 4C). We noted that cancer specific survival and mortality rate after 3 years were slightly different from that within 3 years (Figs 2A, 3A,B and 4C). To investigate this difference thoroughly, we split patients into short-term (<3 years) group and long-term (> = 3years) group. Result showed that neoadjuvant radiation improved short-term survival in both younger and older patients, especially for older patients. However, there was no long-term survival benefit in older group and younger group, especially for younger group (see Supplementary Fig. 3). Additionally, based on 3 th year survival we calculated the 5 year conditional survival which was a kind of more accurate quantification of prognosis for long-term survivors 7 . Result showed that neoadjuvant group has poorer 5 year conditional survival than no-neoadjuvant group (28.43% vs. 30.05%) in younger patients and neoadjuvant group had better 5 year conditional survival than no-neoadjuvant group (26.62% vs. 27.08%) in older patients. The underlying reasons for this long term and short term difference might be variable. Previous study shows that in younger patients who develop the local recurrence die faster than older patients 45 . One possible reason for this might be that in younger patients neoadjuvant radiation group had higher local recurrence rate than no-neoadjuvant radiation group. So in the long term when both groups developed distant metastases and died of disease, neoadjuvant group had poorer survival. However, in older group, although neoadjuvant radiation had better ability to control the local recurrence in the short term, in the long term they had similar local recurrence rate and survival rate.
Additionally, we analyzed the difference of every covariate between two age groups. Fisher's exact test showed that there were significantly different distributions in gender, tumor number, stage, tumor size, lymph nodes resected, histological type and grade between younger group and older group (see Supplementary Table S3), which in same covariates was consistent with previous research 19,23 . Tumors of younger patients tended to be in later stage (Fig. 4D). Since both age and stage were significantly associated with the benefit of neoadjuvant radiation, we further split patients by age and stage to investigate their combination effects. Survival curves were constructed for four subgroups. In younger group (Fig. 5A,B), neoadjuvant radiation did not have survival benefit no matter of the stages, and this trend was worse in stage II (Log-rank P = 0.049, P = 0.163). In older group (Fig. 5C,D), neoadjuvant radiation have survival benefit no matter of the stages, and this trend was better in stage III (Log-rank P < 1.06e-07, P = 0.013). Finally, we calculated the neoadjuvant radiation's HRs in four groups ( Table 2). Result indicated that neoadjuvant radiation played different roles in different sub-groups. Patients in later stage of disease and in older group tended to obtain more survival benefits from the neoadjuvant radiation.

Discussion
In the present study, we used propensity score matching to get a large baseline covariates balanced cohort of rectal cancer, and investigated effectiveness of neoadjuvant radiotherapy on survival. And Cox model confirmed previous reported significant prognostic factors (Fig. 2B, see Supplementary Table S4), such as tumor size 31 , histological type 32 , differentiation 31 , age 6 , number of harvest lymph node 33,34 . In all results, neoadjuvant radiation had cancer specific survival benefit not only in the whole cohort but also in different stage subgroups. Previous studies shows that stage III patients tend to have more survival benefits from neoadjuvant/adjuvant therapy compared with stage II 15,26 , which was consistent with our result. However, neoadjuvant radiotherapy had opposite survival benefit in younger and older groups.
We analyzed the effectiveness of neoadjuvant radiotherapy according to age. Neoadjuvant radiotherapy in younger group showed increasing survival risk. From our disease-specific mortality analysis, neoadjuvant radiotherapy was associated with increase disease-specific mortality starting from 3 years (Fig. 4C). Recently the incidence of CRC tends to be increasing among adults younger than 50 years 19 , and previous studies shows that rectal and distal colon are identified as predilection location in younger CRC patients 19,23 . Age is an important factor when considering the treatment for patients 24 . Therefore they suggests starting rectal cancer screening before age 50 46 . These studies showed that more attention has been paid to younger patients. Our results on neoadjuvant radiotherapy also suggested that younger patients should be distinguished from older patients when making treatment decision.
Since this work was based on population-based cancer registry, there exist some limitations. Population study has limited internal validity compared with randomized controlled trials (RCTs), exists many confounding variables and study design is limited 47 . Many pitfalls exist in population-based state cancer registry data, such as: multiple cancer diagnoses, duplicate reports, reporting delays, misclassification of race/ethnicity, and pitfalls in estimations of cancer incidence rates, etc 48 . So analysis results based on population-based study are prone to multiple biases. In this study, we adopted propensity score matching method to mitigate some potential sources of biases. After matching, sensitivity analysis showed (Fig. 2C) that unmeasured confounders did not influence our conclusions except their hazard ratio higher than 1.3. Such result was similar with previous study 27,41 and indicated the robustness of our results.
Although many covariates had being considered in our analysis, some important clinical factors and prognostic biomarkers were not included due to the lack of information in SEER database, such as: MSI, expression of biomarkers, local recurrence, details of radiotherapy such as dose, field, radiation treatment technique etc. Previous study shows that different age group (< = 50, >50) has different prognostic biomarkers, such as the PRL, RBM3, Wrap53, p53 and DNA status. PRL expression was negatively related with CSS in young CRC patients 6 . Receiving neoadjuvant radiotherapy, strong PRL expression and FXYD-3 expression is unfavorable prognostic in patients 49,50 . Some other genes has been assessed the predictor's role in neoadjuvant radiochemotherapy such as p21, VEGF etc 51 . Besides, Previous research finds MSI is very important good marker of prognosis in stage II/III CRC 52 . A former study showed that MSI could not predict therapeutic response to neoadjuvant radiotherapy 53 . However, this relationship has not fully demonstrated. In younger and older rectal patient groups, there is different percentage of MSI and different studies has different MSI percentage in younger and older rectal patients 54,55 . Besides, in younger patients, 2-7% MSI are associated with Lynch syndrome 56 . Apart from these biomarkers, local recurrence is also an important prognostic factor and clinical indicator for quality of neoadjuvant treatment in rectal cancer. Local recurrence is an painful event for rectal cancer patients and limited success is achieved by further salvage surgery treatment 57 . A study of cohort who underwent R0 or R1 resections showed that patients die faster if diagnosing with local recurrence and local recurrence is the single most important indicator for reduced survival in this study 45 . Neoadjuvant radiation can improve local control for rectal cancer conclusively 15,58 . Different neoadjuvant regimens followed by different surgery has different local control rate, and the standard care for stage II and stage III rectal cancer is neoadjuvant radiotherapy followed by Total Mesorectal Excision (TME), which leads to a decrease in local recurrence rates to 6% 25 . Other study shows that short-course preoperative radiotherapy can reduce local recurrence and can maintain this ability when combined with TME 59 . A randomized study showed that short-course and long-course neoadjuvant radiotherapy are not statistically significant in local recurrence rate 13 .
In future, we would like to do more studies about neoadjuvant radiotherapy. Firstly, results obtained from our population-based study are needed to be validated by more comprehensive and accurate data. Therefore, we expect to analyze data from randomized clinical trials via scientific collaboration. Secondly, we would like to explore the molecular-level differences between younger and older patients before and after neoadjuvant radiation. This may help our understanding of the different effects of neoadjuvant radiotherapy according to age. Thirdly, we will investigate the challenging problem of predicting response to neoadjuvant radiotherapy. We hope find an optimal radiotherapy plan for individual patient by combining clinical factors and molecular markers.
In conclusion, our study provided new insights on the neoadjuvant radiation in rectal cancer, especially for the younger patients; And provided reliable information to guide future knowledge translation of neoadjuvant radiation in younger patients, especially for long term ineffectiveness of neoadjuvant radiation in younger patients. Additionally, our results provided evidence of effectiveness of neoadjuvant radiotherapy for older patients, especially in short term cancer specific survival benefit.  Table 2. HR and P value in different age and stage groups.