Patient Age and Survival After Surgery for Esophageal Cancer

We used data from a Swedish, nationwide, population-based cohort study that has been previously validated and described in detail elsewhere.22,23 Briefly, this study included all patients with esophageal cancer treated with curative intent between 1987 and 2010, with follow-up until November 2014. From the Swedish Cancer Registry, patients with a diagnosis of esophageal cancer (150.0, 150.8, or 150.9) between 1987 and 2010 were identified according to the 7th edition of the International Classification of Diseases (ICD7). Patients with esophageal cancer who underwent esophagectomy were identified from the Swedish Patient Registry. The Swedish Causes of Death Registry was used to provide accurate data about the dates and causes of death. If the diagnosis of esophageal cancer was listed as a cause of death, then mortality was defined as disease-specific. The final source cohort consisted of 1820 patients who underwent curatively intended esophagectomy for esophageal adenocarcinoma or squamous cell carcinoma in Sweden between 1987 and 2010. They were followed up through 2016, thus allowing 5-year follow-up of all participants. After exclusion of patients with missing information on education (n = 69, 3.8%) or tumor stage (n = 14, 0.8%), the final sample included 1737 (95.4%) patients. The data were retrieved from medical records and five national Swedish health and population registries: Cancer Registry, Patient Registry, Cause of Death Registry, Prescribed Drug Registry, and Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA). The linkages of individuals’ data were enabled by the personal identity numbers given to each Swedish resident upon birth or immigration.24 The study was approved by the Regional Ethical Review Board in Stockholm, Sweden.

The study exposure was the patient’s age at the date of surgery. This information was available in all registries and the medical records.

The main outcome was 5-year all-cause mortality, which was selected because the 5-year cut-off reduces the influence of death due to factors unrelated to the esophageal cancer or its treatment, and the all-cause survival is the most complete and accurate assessment of survival. Secondary outcomes were 90-day all-cause mortality and disease-specific 5-year mortality. The choice of 90 days as the short-term mortality cut-off was based on research showing a time shift of postoperative mortality from 30 days to 31–90 days with more recent calendar time.25 The definition of disease-specific mortality was mortality with an esophageal cancer diagnosis as a primary or contributing cause of death after excluding the first 90 days of surgery. The data on all mortality outcomes were retrieved from the Cause of Death Registry, which has over 99% completeness for date and cause of death.26

The confounders were seven variables known to be associated with both the exposure (age) and the outcome (mortality): (1) sex (male or female); (2) attained education level (≤ 9 years, 10–12 years, or > 12 years of formal education); (3) comorbidity (Charlson comorbidity index score 0, 1, or ≥ 2);27 (4) neoadjuvant therapy (typically chemoradiotherapy, no or yes); (5) tumor histology (adenocarcinoma or squamous cell carcinoma); (6) pathological tumor stage (0–II or III–IV); and (7) resection margin status (no tumor involved [R0] or tumor involved [R1 or R2]). Data on sex were retrieved from the medical records and all the registries used. Information on comorbidities was collected from the Patient Registry, which has a high completeness and accuracy of recording of diagnoses,28 while information on neoadjuvant therapy, tumor histology, pathological tumor stage, and resection margin status was obtained from review of all relevant surgery charts and histopathology records from the Swedish hospitals where the esophagectomy had been conducted.22,23

The probability of mortality was computed using a novel flexible parametric model approach29,30 and the results were displayed graphically. The probability of death was defined as the average probability of dying in an infinitely short time interval. This quantity is related to the hazard function through the following mathematical expression: g(t) = 1 − exp[–h(t)], where g(t) is the probability function, h(t) is the hazard function, and exp is the exponential function (see working paper at http://​www.​imm.​ki.​se/​biostatistics/​eventprob/​Working_​paper_​2020.​pdf). This quantity can be readily estimated using general parametric survival models with a user-defined baseline hazard function. Odds ratios (OR) of death as a function of both a linear age variable and a categorical age variable were reported. For comparison, flexible semi-parametric models were used to compute hazard ratios (HR) as a function of age at surgery. All analyses were conducted using the user-written Stata program stpreg available at http://​www.​imm.​ki.​se/​biostatistics/​eventprob.

We first analyzed age as a continuous variable in relation to the mortality outcomes. Several models with different age and time functions, and age and time interactions, were tested and the final model was selected based on the Akaike’s information criterion. Age was introduced into the model as a linear predictor and through restricted cubic splines. Time since surgery was modeled with restricted cubic splines. The interaction between the exposure and time was also tested. Second, we analyzed age in three categories (< 70, 70–74, and ≥ 75 years) in relation to the probability of dying. The cut-off of 70 years of age was used because it is the most common cut-off used in the literature. The cut-off of 75 years of age was introduced to examine associations among older people.

Patients entered the study at the date of surgery and remained in the cohort until the date of death or end of the study (31 December 2016), whichever occurred first. For each of the three mortality outcomes, we analyzed a crude model and an adjusted model. The adjusted model included the seven confounders presented above, using the same categorization. Stratified analyses were carried out for sex, tumor histology, pathological tumor stage, and resection margin status.

Characteristics of the 1737 patients included in the study are described in Table 1. Median age at surgery was 65.6 years (interquartile range 5.6–71.9 years), and the age range was 18–89 years.

The overall 90-day mortality rate was 11.2% (n = 195), and 90% of patients aged below 70 years and 85% of those aged 75 years or older were alive 90 days after surgery (Fig. 1a). Figure 2 shows the changes in probability of dying within 90 days of surgery, depending on age or days after surgery. The probability of 90-day mortality increased from 10% at age 60 years to 60% or above at age 89 years (Fig. 2a). The odds of dying increased by 5% with each year of age (OR 1.05, 95% confidence interval [CI] 1.02–1.07) (Table 2). Patients aged 75 years or older had almost three times the odds of dying compared with patients younger than 70 years of age (OR 2.85, 95% CI 1.68–4.31). There was no statistically significant interaction between age and any of the seven covariates included in the model.

The 5-year all-cause mortality rate was 74.8% (n = 1299). At 5 years after surgery, 27% of patients aged below 70 years and 16% of those aged 75 years or older were alive (Fig. 1b). The association between age and probability of dying is depicted in Fig. 3. The probability of death increased with age (Fig. 3a), but the strength of the association decreased with time after surgery (Fig. 3b). The adjusted odds of 5-year mortality increased by 2% with each year of age (OR 1.02, 95% CI 1.01–1.03) (Table 2). Patients aged 75 years or older had 56% higher odds of dying than those younger than 70 years of age (OR 1.56, 95% CI 1.27–1.92) (Table 2). No interaction term between age and sex, tumor histology, tumor stage, or resection margin status was statistically significant. In the stratified analyses, the strength of the association between age and 5-year all-cause mortality was stronger in patients with clear resection margins (R0) (Table 3). The results remained similar after excluding the first 90 days after surgery (electronic supplementary Table 4 and electronic supplementary Table 5).

The 5-year disease-specific mortality rate was 76.7% (n = 1183). The adjusted odds of disease-specific death increased by 1% for each year of age (OR 1.01, 95% CI 1.01–1.02) (Table 2). The association between age and probability of dying showed a pattern similar to that of the 5-year all-cause mortality (electronic supplementary Fig. 4). There was no difference between the 70–74 years age group and younger patients, while patients aged 75 years or older had 38% higher odds of dying than patients younger than 70 years of age (adjusted OR 1.38, 95% CI 1.09–1.76). There was no statistically significant interaction between age and any of the covariates in the model. In the stratified analyses, older age was associated with higher odds of 5-year disease-specific mortality in participants with clear resection margins (R0) and in patients with early-stage tumor (pathological tumor stage 0, I, or II) [electronic supplementary Table 6].

Electronic supplementary Table 7 reports the powers of the probability of all-cause 90-day and 5-year mortality, which were equal to the ratios of the hazard of dying. Powers and ORs of dying were similar for values of the odds and of the hazards closer to one.