Blood markers of oxidative stress are strongly associated with poorer prognosis in colorectal cancer patients

2.1 Study design and population

Our analysis is based on data from CRC patients who were diagnosed in 2003-2012 and recruited into the DACHS (Darmkrebs: Chancen der Verhütung durch Screening) study. The DACHS study is a population-based case-control study in the Rhine-Neckar region of Germany, with additional regular follow-up of cases. Patients with first time diagnosis of CRC (International Classification of Diseases, 10th Revision [ICD-10], codes C18-C20) and aged 30 years or older and physically able to participate in an interview lasting for about 1 hour were eligible. Patients were recruited from all 22 hospitals providing first-line treatment for CRC in the study region of about 2 million inhabitants. In the recruiting hospitals, eligible patients were informed about the study by their physicians and recruited either during or shortly after their hospital stay. Data from a population-based cancer registry indicate that about half of the eligible patients in the study region were recruited. There was also an age gradient, with higher rates of recruitment of younger patients, in whom comorbidities are often lower and chemo(radio)therapy administration and CRC survival rates are usually higher compared to older patients.^{15, 16} However, the DACHS study did not set an upper age limit, which has merits when aiming to assess the association of oxidative stress with CRC prognosis, as oxidative stress increases with age.¹¹ Further details of the DACHS study have been described elsewhere.^{17, 18} The DACHS study was approved by the ethics committees of the Medical Faculty of Heidelberg University and the state medical boards of Baden-Wuerttemberg and Rhineland-Palatinate. All participants gave written informed consent.

2.2 Inclusion criteria

Our study population comprised all patients diagnosed in 2003-2012 who had data on either d-ROMs or TTL (n = 3730, 90%; Figure S1). Patients with poor hemolytic, icteric or lipemic serum samples and very low d-ROMs (<50 U.Carr) or TTL (<50 μmol/L) according to the manufacturer of the assays (n = 28), those who were not operated on for CRC (n = 50), those whose blood samples were taken within 3 days of CRC surgery or within 7 days of chemo(radio)therapy initiation (n = 141) and those with no information on tumor stage and other variables of interest (n = 150) were excluded. The cut-offs for CRC surgery and chemo(radio)therapy initiation were chosen based on reasoning, as surgery and chemo(radio)therapy are likely to have an immediate effect on d-ROMs and TTL levels. A total of 3361 patients were included in the analysis.

2.3 Data collection

At baseline (mostly during or shortly after hospital stay for CRC surgery), trained interviewers conducted personal interviews with the participants to collect information on lifestyle factors and medical history, using a standardized questionnaire. Tumor characteristics (eg, stage and subsite) and ICD-10 codes for comorbidities that were diagnosed either prior to or at the time of CRC diagnosis were abstracted from hospital records. We used the Charlson comorbidity index (CCI),¹⁹ adapted by Deyo et al,²⁰ to quantify overall comorbidity as described elsewhere.^{15, 21} Patients were grouped into four groups, namely, CCI scores 0 (no comorbidity), 1, 2 or 3+ (severe comorbidity). Vital status and cause of death were ascertained from population registries and public health authorities about 3, 5 and 10 years after CRC diagnosis. About 3 years after CRC diagnosis, detailed information on treatment was collected from medical records and from questionnaires sent to patients’ oncologists.

2.4 Ascertainment of oxidative stress biomarkers

At baseline, blood samples were taken from patients (median time from diagnosis to blood sampling was 63 days, interquartile range [IQR], 23-270 days), from which serum aliquots were obtained and stored at −80°C until analysis for various measures, including d-ROMs and TTL. Whenever possible, patient recruitment and sample collection was prior to or shortly after CRC surgery and prior to chemo(radio)therapy initiation. Of patients receiving chemo(radio)therapy, approximately 40% had their samples collected prior to initiation of this treatment. Median frozen time was 7.9 (IQR, 5.8-10) years. A previous study suggests that d-ROMs and TTL have a good long-term stability under storage conditions of −80°C.²²

Samples were analyzed in two batches in the same laboratory and under similar conditions (70% in April, 2016 and 30% in June, 2016). Standardized assays that are used to measure d-ROMs and TTL (d-ROMs and SHp assay, respectively, both from Diacron, Grosseto, Italy) were adapted to an auto-analyzer (LX20-Pro, Beckman-Coulter, Woerden, The Netherlands) at the Laboratory for Health Protection Research (Bilthoven, The Netherlands). Of samples analyzed in the first 8 days, quality assessment showed comparable mean values, with no evidence of day-to-day variability. In brief, the d-ROMs assay evaluates the total hydroperoxide concentration in Carratelli Units (U.Carr). Each U.Carr is equivalent to 0.08 mg of hydrogen peroxide per 100 mL in the sample.²³ The higher the U.Carr, the higher the level of oxidative stress. The SHp test is a spectrophotometric test which estimates the concentration of free thiol groups (eg, sulfhydryl groups) in serum in μmol/L using information on color intensity. Higher values indicate lower levels of oxidative stress.

2.5 Statistical analysis

The mean values of the biomarkers, especially TTL, differed by year of blood sampling. To address this, we calculated a z-score for each patient by subtracting the mean of each year of blood sampling from the value recorded for each patient and dividing by the SD of that year.²⁴ This standardized the data such that the mean value of each year's sample was zero and the SD was one. We assessed the distributions of z scores of d-ROMs and TTL by baseline characteristics. ANOVA tests were used to test the mean differences, by adjusting for age and sex.

The associations of z scores of d-ROMs and TTL (in quintiles) with overall survival (OS, mortality from any cause) and disease-specific survival (DSS, mortality from CRC) were investigated using Cox proportional hazards regression. d-ROMs and TTL violated the proportional hazards assumption, as their associations with survival outcomes became weaker after 6 years of follow-up. To address this, we censored patients at 6 years of follow-up (median follow-up time), as appropriate. Time was calculated from CRC diagnosis to the respective endpoints or end of follow-up, whichever occurred first. We also investigated the association of the TTL to d-ROMs ratio with OS and DSS. The TTL to d-ROMs ratio might reflect a balance between antioxidant and oxidant capacities and has recently been found to predict all-cause mortality more strongly than TTL or d-ROMs alone in patients with type II diabetes.²⁵

Two adjustment levels were applied: (a) adjustment for sex, age, tumor stage (Union for International Cancer Control, I-IV),²⁶ tumor site, year of diagnosis, education level, smoking status, body mass index (BMI) at diagnosis, lifetime physical activity and alcohol consumption, vegetable and fruit consumption and time of blood draw with respect to CRC surgery and chemo(radio)therapy initiation, and (b) additional adjustment for comorbidity score. Results from model (b) were reported as main results and were also used to assess to what extent comorbidities explain the association between oxidative stress and CRC prognosis. All covariates were added to the models as categorical variables, as listed in Table 1 and Figures S2 andS3. We checked the proportional hazards assumption for all covariates, by assessing whether their interaction with follow-up time was statistically significant. Time-dependent covariates (interaction terms of BMI, year of diagnosis, and tumor site with follow-up time) were also added to the models because of violation of the proportional hazards assumption. We also accounted for left truncation by including “delayed entry time” in the models. In sensitivity analysis, we used the Fine and Gray method²⁷ to account for competing events (mortality from causes other than CRC) in the association of the biomarkers with DSS. Also, in a subsample of patients diagnosed in 2003 to 2010 who were selected for tumor characterization (n = 1876, 77%), we furthermore assessed whether and to what extent the association of oxidative stress is partly explained or modified by microsatellite instability (MSI), a prognostic factor for CRC.²⁸

Subgroup analyses according to sex, age, comorbidity, smoking status, tumor stage and subsite, MSI status, chemo(radio)therapy use, time of blood sampling (tertiles of d-ROMs and TTL instead of quintiles were used because of small numbers) were performed. Here, we used backward selection to select covariates with P < .6 (defined a priori) for the models, but age, sex and stage were forced into the models. Potential variations of associations across subgroups defined by these variables were assessed with statistical tests for interaction. Stage-specific results were additionally illustrated with adjusted survival curves. We furthermore assessed potential linear associations of the biomarkers with all-cause mortality according tumor stage, using restricted cubic spline functions.²⁹ Knots were placed at the 25th, 50th and 75th percentiles. Sensitivity analyses were also conducted using different knot positions (eg, at 5th, 65th and 95th percentiles).

Finally, we assessed whether and to what extent adding d-ROMs and TTL to models that included age, sex, tumor stage and tumor subsite improve the prediction of CRC prognosis, overall and by tumor stage using the Harrell's concordance index (C-index). Here, we created 1000 bootstrapped samples and trained the models in 2/3 of each of the samples (training set) and evaluated the predictive accuracy of the regression coefficients in the remaining 1/3 of the samples (validation set). The samples were split into training and validation sets using random sampling.

The differences in the C-indexes of the two prediction models were tested for statistical significance using the “compareC” package in R (version 3.6.0). All other analyses were conducted with the SAS software, version 9.4 (SAS Institute, Cary, North Carolina). Statistical tests were two-sided, with a significance level of 5%.

3.1 Characteristics of the study population

A total of 3361 patients were included in the analysis (Figure S1). The median age was 69 (IQR, 62-76) years. About 24%, 31%, 32% and 13% of the patients were in stage I, II, III and IV, respectively. Of the analytic sample, 3237 and 3254 patients had data on d-ROMs and TTL, respectively. There was a weak negative correlation between d-ROMs and TTL (r = −.13; P < .001). Table 1 shows the distribution of mean z-scores of d-ROMs and TTL by baseline characteristics. d-ROMs were particularly high in women, obese patients, current smokers, those not consuming alcohol and stage IV patients. TTL was especially low in older and obese patients, those with severe comorbidities, and stage IV patients.

d-ROMs and TTL levels also differed by time of blood sampling (Figures S2 and S3). For d-ROMs, mean z-score was highest among patients whose blood samples were taken within 2 to 8 weeks of CRC surgery. Among patients who received chemo(radio)therapy, d-ROMs levels were higher when blood samples were taken prior to rather than after initiation of such therapy. For TTL, levels were lowest within 2 weeks and highest more than 8 weeks from CRC surgery (mean z scores <−0.46 and >0.38, respectively). TTL was lowest when samples were taken before chemo(radio)therapy initiation (−0.32) and highest among patients with samples taken 5+ months after chemo(radio)therapy initiation (>0.31). Regarding time of blood sampling from CRC diagnosis, mean values of d-ROMs and TTL levels were similar to those reported for time of blood collection defined by CRC surgery (data not shown).

3.2 Associations of oxidative stress biomarkers with CRC prognosis

During a follow-up period of 6 years, 1078 (32%) deaths occurred, of which 708 (66%) were due to CRC. Table 2 shows hazard ratios (HRs) for the associations of d-ROMs, TTL and their ratio with CRC prognosis. In patients of all stages, those in the highest d-ROMs (vs lowest) quintile had 27% increased all-cause mortality, but this association was slightly attenuated and lost statistical significance after adjusting for comorbidities (HR = 1.21, 95% CI = 0.99-1.48). Higher TTL was associated with substantially lower mortality. HRs for the second, third, fourth, and fifth quintiles, compared to the lowest quintile, were 0.92, 0.66, 0.58 and 0.48 for OS (P < .001) and 0.87, 0.58, 0.58 and 0.52 for DSS (P < .001). Even stronger associations were seen for the TTL to d-ROMs ratio. HRs for the second, third, fourth and fifth quintiles, compared to the lowest quintile, were 0.79, 0.59, 0.55 and 0.42 for OS (P < .001) and 0.72, 0.55, 0.56 and 0.49 for DSS (P < .001). In the competing risk models, the associations of higher levels of TTL and the TTL to d-ROMs ratio with lower CRC mortality were only slightly attenuated. Further adjustment for MSI status did not affect the observed associations (Table S1).

3.3 Subgroup analyses by tumor stage and subsite

In stage-specific analyses, we observed significant associations between higher d-ROMs and poorer survival in stages I-II and IV but not in stage III (Table 3 and Figure 1). Compared to the lowest tertile, stage I-II (HR = 1.38, 95% CI = 1.04-1.83) and IV patients (HR = 1.52, 95% CI = 1.14-2.04) in the highest tertile had poorer OS. In stage IV patients, those in the highest (vs lowest) tertile had much poorer DSS (HR = 1.61, 95% CI = 1.20-2.17). For TTL (Table 3 and Figure 2), strong inverse associations of TTL with both all-cause and CRC mortality were observed within all stages, but the associations were particularly pronounced for stages I-II and IV (P_interaction = .070 for OS). Even slightly stronger associations of higher TTL to d-ROMs ratio with lower all-cause and CRC mortality were seen within all stages, with particularly pronounced associations in stage IV patients (P_interaction = .013 for OS; Table 3 and Figure 3).

Figures S4 and S6 show stage-specific assessment of linear associations of d-ROMs and TTL with OS. For d-ROMs, nonlinear associations were observed in stages I-II and IV, with a steep increase in mortality for z-scores >0, which was especially strong for stage IV patients (Figure S4). A U-shaped association was observed in stage III, with a much higher mortality in patients having z scores <−1.0 or >1.0. A strong monotonic association between higher TTL and lower mortality was seen in all stages (Figure S5). For the TTL to d-ROMs ratio, nonlinear associations with OS were observed, with a much lower survival in patients having z-scores <0 (Figure S6). No significant interactions between the oxidative stress biomarkers and tumor subsite for the investigated survival outcomes were observed (data not shown).

3.4 Subgroup analyses by sex, age, comorbidity, smoking status, MSI status and chemo(radio)therapy

There was a significant interaction between d-ROMs and comorbidity for OS (Table S2). For example, higher d-ROMs (highest vs lowest tertile) were associated with poorer OS in patients without comorbidities (HR = 1.45, 95% CI = 1.15-1.82), whereas no associations were seen in patients with comorbidities (P_interaction = .037). For TTL, similarly higher OS with increasing TTL was seen across genders, comorbidity levels and smoking categories. The associations of TTL with survival were, however, restricted to patients aged <75 years (P_interaction = .001). Subgroup analyses by MSI status (Table S3) showed particularly pronounced associations of higher d-ROMs levels with lower survival among MSI-high patients, with less clear patterns of TTL levels with survival among MSI-high patients. No significant interactions between the oxidative stress biomarkers and chemo(radio)therapy use (yes/no) for the investigated survival outcomes were observed in stage II-IV or II-III patients (data not shown).

3.5 Subgroup analyses by time of blood sampling

Comparable associations of d-ROMs and TTL with OS were seen across strata defined by time of blood sampling with respect to CRC surgery and chemo(radio)therapy initiation (Table S4). However, the associations of TTL with OS seemed weaker in recipients of chemo(radio)therapy whose blood samples were taken within 5 months of initiation of such treatment (P_interaction = .001). Also, an inverse association between d-ROMs and all-cause mortality was seen in this patient group. But among patients receiving chemo(radio)therapy, there was no difference in all-cause (HR = 0.99, 95% CI = 0.76-1.29) or CRC mortality (HR = 1.04, 95% CI = 0.78-1.39) between patients whose samples were taken after and before initiation of such treatment (data not shown).

3.6 Added value of d-ROMs and TTL in the prediction of CRC prognosis

Table 4 shows the C-indexes of models that included age, sex, tumor stage and site and those that additionally included d-ROMs and TTL. Results showed improvement in the prediction of CRC prognosis by the addition of d-ROMs and TTL in both the training and validation sets. Results from the validation set were very similar to those from the training set and the whole sample. In the whole study sample, the C-indexes of models with and without d-ROMs and TTL were 0.772 and 0.755 (P < .001) for OS and 0.832 and 0.823 (P < .001) for DSS, respectively. Stage-specific analyses similarly demonstrated superiority of the models with to those without d-ROMs and TTL in all stages and for both OS and DSS, with particularly pronounced improvements in stage IV patients. In stage IV patients, the C-indexes of models with and without d-ROMs and TTL were 0.615 and 0.566 (P = .001) for OS and 0.608 and 0.557 (P = .002) for DSS, respectively.