Does diabetes affect paraneoplastic thrombocytosis in colorectal cancer?

2.1 Patients and study design

A prospective, real-life observational cohort study was carried out. A total of 108 patients diagnosed with CRC and 166 voluntary non-CRC subjects were enrolled for the study between 2017 and 2019. CRC patients attended at both the Department of Internal Medicine and Hematology, Semmelweis University, Budapest and the Department of General Surgery, Szent Imre University Teaching Hospital, Budapest. Written informed consent was collected from all study participants before performing any study-specific procedures. Exclusion criteria included age <18 years, any previous malignancies, known inflammatory bowel- and/or chronic kidney- and/or systemic autoimmune- and/or inadequately controlled thyroid- and/or hematologic- and/or any mental diseases, the usage of erythropoiesis-stimulating agents and/or recent blood transfusion, and patients with an Eastern Cooperative Oncology Group (ECOG) performance status >2.

Voluntary non-CRC subjects consisted of healthy young controls, older controls, and T2DM patients. Rationale of the inclusion of healthy young subjects was that some of the measured parameters have no reference values and that the older volunteers included both completely healthy individuals and nonmetabolic disease patients (e.g., hypertension). T2DM patients attended at the Metabolic Outpatient Clinic of the Department of Internal Medicine and Hematology, Semmelweis University, Budapest. In addition to the exclusion criteria described by CRC patients above, voluntary nondiseased subjects were excluded in the presence of prediabetes or any other metabolic disorders.

2.2 Clinical and laboratory data measurements

Anamnestic data including comorbidities and recent medications, body weight, and height were collected, and fasting blood samples were drawn according to the following protocol: (1) at the diagnosis of CRC, before any oncological treatments or the surgical resection of the primary tumor, (2) at least 6 weeks, (3) 6 months, and (4) 12 months after the tumor removal surgery. At the time of postoperative measurements, oncological treatments were stopped/paused only as part of the patients’ standard of oncological care, and sampling was performed only at least 6 weeks after the last treatment due to the known platelet influencing effects of several cytotoxic regimens [21,22]. If it was not feasible to stop/pause the oncological treatment, in the best interest of the patient, the visit was omitted. Complete blood count, aspartate- and alanine transaminase, gamma-glutamyl transferase, plasma glucose, and creatinine were measured at the Central Laboratory of Semmelweis University and at the Central Laboratory of Szent Imre University Teaching Hospital. Estimated glomerular filtration rate was calculated using the Chronic Kidney Disease-Epidemiology Collaboration equations [23]. Plasma interleukin-6 and thrombopoietin levels were measured using the ELECSYS^® Interleukin-6 electrochemiluminescence immunoassay (ECLIA) (Roche Diagnostics GmbH, Mannheim, Germany) and the Human Thrombopoietin Quantikine^® enzyme-linked immunosorbent assay (ELISA) (catalog number: DTP00B, R&D Systems, Minneapolis, MN, USA) kits, respectively. Plasma thrombopoietin level was measured from platelet-poor plasma, as per manufacturer description.

Side of CRC was described as right sided if the tumor was originating from cecum, ascending colon, and proximal two-third of the transverse colon and left sided if originating from the distal one-third of the transverse colon, descending colon, sigmoid colon, and rectum [24]. Staging was given by histopathological examination of surgical specimens and imaging studies; the American Joint Committee on Cancer grouping was used [25]. The usage of biological agents was recorded as a dummy variable and chemotherapy was grouped as adjuvant if no metastasis and first-line, second-line, etc., if metastasis was present. The overall survival of patients was defined as the length of time from the date of CRC diagnosis until death. Follow-up of cancer patients was terminated on January 31, 2021, patients alive at this time point were right censored.

2.3 Statistical analysis

Statistical analysis was performed with R version 4.0.5 [26]. Wilcoxon–Mann–Whitney U test, Fisher’s exact test, Kruskal–Wallis test with p-value corrected pairwise Wilcoxon–Mann–Whitney U tests as post-hoc, and Spearman rank correlation were used. To detect the changes of various parameters in time, linear mixed effect models were used (R-package nlme [27]). Survival of patients was analyzed for both preoperative and longitudinal data with Cox regression models and Bayesian univariate and multivariate joint survival models (R-package rstanarm [28]), respectively. p < 0.05 was considered as statistically significant, and p-values were corrected with the Holm method [29] for the problem of multiple comparisons. Compared to the “conventional” frequentist methods, Bayesian methods give only a probability distribution for the investigated parameter but no p-values. Interpretation of joint model results was as follows. If the Bayesian equivalent to the frequentist confidence interval (CI), the 95% credible interval contained the hazard ratio (HR) of 1, the model was considered as clinically not relevant. However, if the 95% credible interval was less than or more than HR = 1, the effect of the parameter was considered as a good or bad sign of patient survival, respectively. Results were expressed as mean ± standard deviation and as the number of observations (percentage) for continuous and count data, respectively. Naïve Kaplan–Meier and longitudinal survival curves were drawn with the R-package survminer (Kassambara, Kosinski, and Biecek, version 0.4.9, 2021) and the built-in methods of rstanarm, respectively.

1. Informed consent: Informed consent has been obtained from all individuals included in this study.

2. Ethical approval: The study was approved by the Regional and Institutional Committee of Science and Research Ethics, Semmelweis University (SE TUKEB 21-12/1994, approval date of latest modification: February 10, 2017), by the Institutional Review Board of Szent Imre University Teaching Hospital (SZIK IKEB 5/2017), and by the Committee of Science and Research Ethics, Hungarian Medical Research Council (ETT TUKEB 8951-3/2015/EKU).

3.1 Baseline measurement of paraneoplastic thrombocytosis-related parameters

Plasma level of paraneoplastic thrombocytosis parameters of patients was compared to those of all control groups at the time of CRC diagnosis. Highest platelet counts were observed within the two tumor groups: platelet count of CRC patients was significantly higher than those of within the T2DM group (p = 0.0369), whereas the platelet count of the CRC + T2DM group was significantly higher than all of the control groups (p = 0.0369 vs Young Control and Control, p = 0.0278 vs T2DM; Figure 1a). Lowest plasma interleukin-6 levels were observed within Young controls (p < 0.0010 vs all other cohorts). Subjects of the Control and T2DM groups had similar interleukin-6 levels; and interleukin-6 was significantly higher in both tumor groups, compared to all of those observed in control groups (p < 0.0001 CRC vs Control and T2DM, p = 0.0011 CRC + T2DM vs Control, and p = 0.0069 CRC + T2DM vs T2DM; Figure 1b). Thrombopoietin level was basically the same in the Young control and Control groups, and a separate cluster was formed by the remaining three groups. The highest thrombopoietin levels were observed in the CRC + T2DM group (Figure 1c). No further difference could have been justified in any of the study groups if they were further subdivided by the usage of platelet aggregation inhibition therapy, antidiabetic drugs, or in the presence of any diabetic complications.

Figure 1

Comparison of paraneoplastic thrombocytosis-related parameters within the five study groups. (a) The platelet count was significantly higher in CRC patients with T2DM, compared to those in all control groups, and it also differed between the CRC and T2DM groups. (b) The plasma interleukin-6 level was significantly higher in all tumor patients¹, whereas (c) the plasma thrombopoietin level was significantly lower within the T2DM and CRC, and significantly higher in the CRC + T2DM groups, compared to those of control subjects. Note: p-value correction was used for all between-groups comparisons. ¹Due to better visibility, six and one outlier(s) over 40 pg/mL were not shown in the CRC and CRC + T2DM groups; maximum values were 220.20 and 77.19 pg/mL, respectively.

Highest platelet counts were found if interleukin-6 was high as well (CRC: Spearman ρ: +0.34, p = 0.0017; CRC + T2DM: p = 0.0786; CRC groups combined: Spearman ρ: +0.32, p = 0.0009). Correlation between platelet count and thrombopoietin levels was only significant in the CRC group (Spearman ρ: –0.24, p = 0.0332), whereas no (p = 0.8486) and marginal association (p = 0.0643) was found in the CRC + T2DM and in the two tumor groups combined, respectively. No correlation was found between interleukin-6 and thrombopoietin levels (CRC: p = 0.4279, CRC + T2DM: p = 0.9921, tumor groups combined: p = 0.5383). No correlation was found between the thrombocytosis-related parameters and the duration of T2DM or the preoperative level of HbA_1C.

3.2 Changes in the parameters of paraneoplastic thrombocytosis with the course of CRC

For the 108 CRC subjects, a total of 215 measurements were available. 108, 48, 37, and 22 preoperative, postoperative, 6-month, and 12-month measurements were available, respectively. Significant decrease in later measurements occurred due to the death of patients, disease progression resulting in higher ECOG performance status and patient’s unavailability to attend at later visits, the need to initiate chemotherapy earlier than the postoperative visit window, or continuous chemotherapy without drug holiday after the second study visit. Due to the decreasing number of follow-ups, a more robust statistical method not sensitive to the loss of follow-up had to be chosen; therefore, it was investigated via age-corrected and stage-corrected linear mixed effect interaction models as to how T2DM and worse clinical outcome (death) affect the changes of platelet count, interleukin-6, and thrombopoietin levels with the course of the disease. Average survival time: 16.96 ± 11.43 months, all within the first 3 years after CRC diagnosis. Diabetes had no significant effect on any of the changes (platelet count: p = 0.1190; interleukin-6: p = 0.5571; thrombopoietin: p = 0.3062, Figure 2a, c, and e). Platelet counts of all patients decreased, but among those patients who died during the time of the study, the average baseline platelet count was significantly higher (273.70 vs 353.00 × 10⁹/L; p = 0.0042) and a faster decrease could have been seen within the first 12 months after the tumor removal surgery compared to those of who survived (Figure 2b). Similarly, increased baseline interleukin-6 levels (5.76 vs 27.42 pg/mL; p < 0.0001) and marginally faster decreasing levels (p = 0.0613) were observed in deceased patients over time, whereas in those who survived the plasma interleukin-6 level was constant during our observation (p = 0.1273, Figure 2d). The initial thrombopoietin level did not differ between surviving and deceased patients (p = 0.5747), and its change was not affected by the worse clinical outcome (p = 0.5940, Figure 2f). Usage of radiotherapy and chemotherapy did not affect any of the response variables, if included within any of the models.

Figure 2

Change in the parameters of paraneoplastic thrombocytosis with the course of the disease. No difference was found in any of the parameters (a, c, e) when stratified with the presence of T2DM. (b) Platelet count and (d) plasma interleukin-6 were significantly higher at the time of CRC diagnosis in those patients who died during the observation period. Within the first year after tumor removal surgery, a significantly faster platelet count decrease was found in those patients who died (b), whereas the significant difference in plasma interleukin-6 levels remained throughout the whole study between survivors and those patients who died (d). No statistical difference was found in plasma thrombopoietin levels of survivors and nonsurvivors (p = 0.5940, f).

3.3 Survival analysis of paraneoplastic thrombocytosis-related parameters and T2DM

Thirty of the 108 patients (27.8%) died during the study, from which 20 and 10 belonged to the CRC and CRC + T2DM group, respectively. Patients were followed up no later than January 31, 2021. The survival analysis was performed on both the preoperative (single-time) and longitudinal data. Despite the higher occurrence of death (24 vs 40%) within the diabetic tumor group and the difference that appears to be significant on the naïve Kaplan–Meier figure (Figure 3), the univariate Cox model of preoperative data suggested that T2DM had no effect on patient survival (p = 0.1450). A higher preoperative platelet count (HR: 1.0026, 95% CI: 1.0010–1.0050, p = 0.0052) could be considered as a poor prognostic sign. Interleukin-6 and thrombopoietin levels had no significant univariate effect; however, if combined with T2DM in a multivariate model, higher interleukin-6 levels had marginal effect (HR: 1.0007, 95% CI: 0.9995–1.0140, p = 0.0692) on patient survival. Similar to the univariate model, higher platelet counts had the same significant effect in the multivariate model as well (HR: 1.0028, 95% CI: 1.0009–1.0050, p = 0.0043). T2DM did not have any effect in either multivariate model. A subgroup analysis within the CRC + T2DM group only revealed that neither the duration of diabetes (p = 0.5590), the use of any antidiabetics (p = 0.2620), the presence of any diabetic complications (p = 0.2860), nor the preoperative level of HbA_1C (p = 0.5370) affected patient survival.

Figure 3

Naïve Kaplan–Meier curves of CRC patients with and without T2DM. It should be noted that although the two curves do not appear to be similar, neither Cox regression (p = 0.1450), nor log-rank test (p = 0.0908) could justify a statistical difference.

To analyze the effect of paraneoplastic thrombocytosis-related parameter changes in time, Bayesian joint models were used. First, three univariate joint models (a single parameter is analyzed within the longitudinal submodel) were constructed: the survival effect of the changes of platelet count, interleukin-6, or thrombopoietin over time was included in the longitudinal submodel, and T2DM was included in the survival submodel. An additional multivariate joint model was also constructed, where all three paraneoplastic thrombocytosis parameters were included within the longitudinal submodel, and no change was applied to the survival submodel. Interpretation of the clinical significance of parameters on patient survival was assessed as described in Section 2.

Based on the result of the univariate joint models, after the surgical removal of the primary tumor, higher platelet count and plasma interleukin-6 level is a sign of poorer survival (Figure 4). Thrombopoietin and diabetes did not affect the survival of patients in any of the univariate models. Multivariate joint model results suggested that interleukin-6 had the strongest effect on patient survival, whereas the platelet count was marginal; thrombopoietin and diabetes had no clinically relevant effect, similar to those observed in univariate models (Figures 4 and 5).

Figure 4

Longitudinal and survival predictions of the Bayesian joint models. No difference was found between CRC patients with or without T2DM. Increasing platelet count and plasma interleukin-6 level were associated with a higher risk of shorter survival times (left), whereas no change or a slow decline in these parameters is a good prognostic sign (right). Thrombopoietin levels had no clinically relevant effect neither in univariate nor in multivariate survival models.

Figure 5

Results of the Bayesian joint survival models. Platelet count and/or plasma interleukin-6 level increasing over time is a poor prognostic factor, whereas diabetes and plasma thrombopoietin changes did not affect patient survival (Note: credible interval is the Bayesian equivalent of the frequentists’ CI.).

4.1 Limitations

Limitations of the current study were the small sample size and the heterogeneity of CRC population – the latter may be also compensated with a larger sample size. The proportion of T2DM patients within the tumor cohort corresponded to the healthy Hungarian population (approximately every fourth person, over 60 years of age [8]). Due to patients’ decision, a large number of potential subjects did not agree to be included in the study. Despite the low number of cases, it is important to emphasize that the presence of paraneoplastic thrombocytosis could have been already detected at such a low number of cases, showing its significance in CRC. The low number of T2DM in those of CRC patients allowed us only to pinpoint tendentious differences in several parameters.