Lower circulating platelet counts and antiplatelet therapy independently predict better outcomes in patients with head and neck squamous cell carcinoma

The Cancer Registry at Hollings Cancer Center, Medical University of South Carolina (MUSC) was used to identify all cases of HNSCC (squamous cell cancers of the oral cavity, pharynx and larynx). The study population was comprised of histologically confirmed cases diagnosed between January 1, 2000 and June 30, 2012. After excluding patients diagnosed with a second primary cancer between January 1, 1993 (the earliest date recorded) and June 30, 2013, there were 1376 HNSCC patients available for the study. Patients without recorded platelet counts were also excluded, resulting in a total of 1051 patients available for analysis. The follow-up period ranged between 2 weeks and 156 months, with a median 25.7 months.

The MUSC Institutional Review Board approved all study activities. For each case, we abstracted data on demographic characteristics, clinical and pathological variables at diagnosis, treatment received and outcome using two different data sources: the Hollings Cancer Center (HCC) cancer registry and the MUSC Clinical Data Warehouse (CDW). The registry is part of a state mandated data system that collects cancer incidence on all cases in South Carolina. The CDW is a single, secure and integrated database extracted from the MUSC OACIS Clinical Data Repository, which includes patient demographics, ICD-coded diagnoses, ICD-coded procedures, medications and laboratory test results. These databases were subsequently linked, in a blinded fashion, through an honest broker at the CDW, and entered into a secured study database.

Independent variables obtained from the CDW included socio-demographic characteristics (age at diagnosis, sex and race); pre-treatment platelet count, lifestyle factors including smoking status (never, former and current) and alcohol use (never, former and current); and use of platelet-interfering medications, namely NSAIDs or antiplatelet medications, after diagnosis. NSAIDs included aspirin, celecoxib, rofecoxib, valdecoxib, diclofenac, etodolac, ibuprofen, indomethacin, ketorolac, meloxicam, nabumetone, naproxen, oxaprozin, sulindac and choline magnesium trisalicylate. Antiplatelet medications included cilostazol, clopidogrel, prasugrel, ticlopidine, abciximab, eptifibatide, tirofiban, anagrelide, dipyridamole, dipyridamole and aspirin. Intake of medications was based on the inpatient records of the CDW. Tumor-related variables obtained from the registry included tumor grade (well differentiated, moderately differentiated, poorly differentiated/undifferentiated); location of the primary tumor (oral cavity, pharynx, larynx); TNM stage (I, II, III, IV); and receipt of all first-line therapies (chemotherapy, surgery, radiation and/or other). Because there were only 5 cases with undifferentiated grade, we combined these with poorly differentiated grade into a single category for the purposes of analysis.

We evaluated the association between the 5 platelet categories and survival by fitting univariate Cox Proportional Hazards (CPH) regression models. We also analyzed platelets as a continuous variable within the normal range. We chose to restrict the analysis to the normal range because the relationship between platelets and death was non-linear (u-shaped curve) in the lower tail of the distribution. To control for potential confounding variables, we performed univariate CPH for all other variables in Table1 (i.e., age, sex, race, smoking status, alcohol use, treatment, tumor grade, location and clinical stage). Multivariable CPH models were then fit for the platelet variable and all other variables that were found to have a p-value of < 0.20 in the univariate analysis. Variables were retained in the final CPR model if they were significant at p < 0.05. Additionally, for each model, we examined two-way interactions between platelet measures and variables that were significant in the multivariable model. Interactions were retained in the final model if they were significant at p < 0.05.

The methodology described above was used to examine the relationship between post-diagnosis use of NSAIDs/antiplatelets and survival. The association between any post-diagnosis NSAIDs/antiplatelet medication use (Yes or No) and survival was assessed while adjusting for age, race, smoking, treatment, tumor site, tumor grade, platelet level (continuous, or platelet classes) and clinical stage. We also examined the association of medication use and survival while excluding those with pre-diagnostic thrombocytopenia (who would not be expected to benefit from antiplatelet medications). The interaction between post-diagnostic NSAIDs/antiplatelet use and platelet classification survival was also evaluated.

Age inversely correlated with platelet counts, where patients in the low normal and low categories were slightly older than those in the other groups (Table1). The majority of patients were males, but females were enriched in the higher platelet groups (mid-normal, high normal and high) (Table1). In addition, African American patients were more represented in the thrombocytosis group where they comprised more than half of that population. Current smoking and alcohol consumption in turn were significantly associated with higher platelet counts (Table1). Intriguingly, compared to those with stage I-III disease, higher platelet counts were observed in stage IV HNSCC (Table1), although platelet levels did not significantly correlate with tumor grade or anatomic site (Table1). No significant association was observed between intake of antiplatelet medications/NSAIDs and the platelet count. Finally, the proportion of deceased patients was higher in the high normal and thrombocytosis groups.

Compared to those in the mid-normal category, patients in the high normal (adjusted HR 2.20; 95% CI 1.58-3.05) and thrombocytosis (adjusted HR 2.37; 95% CI 1.60-3.50) groups had more than two times higher death rates (Table2 and Figure1). The results for patients with platelet counts below the mid-normal category also trended toward poorer survival with adjusted hazard ratios of 1.25 (95% CI 0.93-1.68) in the low normal and 1.50 (95% CI 0.96-2.35) in the low categories, although these differences were not statistically significant.

Strikingly, when platelet count was modeled as a continuous variable within the normal range, an increase of platelets by 100,000/μL resulted in a 32% increase in the hazard ratio (p = 0.001) in the unadjusted model. The increase in hazard was 30% (p = 0.01) after controlling for age, race, treatment, tumor site, tumor grade, smoking and stage.

The univariate association between platelet count and risk of death was assessed for three tumor sites (oral, pharynx, and larynx). Patients with thrombocytosis compared to those in the mid-normal range had a HR of 4.35 (95% CI 2.75-6.87) for oral cancer, 2.41 (95% CI 1.32-4.44) for pharyngeal cancer, and 1.48 (95% 0.78-2.80) for laryngeal cancer.

Patients who were prescribed antiplatelet medications or NSAIDs after the date of diagnosis had a reduced risk of death compared to the other patients [HR 0.76, 95% CI 0.58- 0.99] after controlling for age, race, stage, grade, tumor site, treatment, smoking status, and platelet count (continuous) (Table3 and Figure2). The HR for medication use and death rate was 0.80 (95% CI 0.61-1.05) when we adjusted for age, race, stage, grade, tumor site, treatment, smoking status, and platelet classification. The association between the use of antiplatelet medications and death was stronger, however, when patients with thrombocytopenia were excluded: HR 0.73 (95 CI 0.55-0.98; p = 0.04). The association between medication intake and survival was assessed within each of the five platelet categories. Within the normal platelet range, intake of NSAIDs/antiplatelet conferred variable HRs in the protective direction, none of which was statistically significant. However, the protective association with intake of antiplatelet medications was strongest in the thrombocytosis group compared to those not taking medications [HR 0.42, 95% CI 0.17-1.05), p = 0.06] (Table3).