A Prediction Model for Chemotherapy-Induced Thrombocytopenia Based on Real-World Data and a Close Relationship Between AST/ALT Ratio and Platelet Count in Patients with Solid Tumors

Introduction

Chemotherapy-induced thrombocytopenia (CIT), a most common complication in cancer patients receiving myelosuppressive chemotherapy, emerges as the peripheral platelet count falls below 100×10⁹/L, with or without bleeding.¹ In the Common Terminology Criteria for Adverse Events version 4.03 (CTCAE 4.03) established by the National Institute of Health (NIH), thrombocytopenia refers to a decrease in the number of platelets (PLT) in the blood specimen, involving four grades: grade 1, PLT count < low limit of normal (LLN) of 75,000/mm³ or 75.0×10⁹/L; grade 2, PLT count <75,000–50,000/mm³ or <75.0–50.0×10⁹/L; grade 3, PLT count <50,000–25,000/mm³ or <50.0–25.0×10⁹/L; grade 4, PLT count <25,000/mm³ or <25.0×10⁹/L. Prior studies estimated that more than 30% of patients with solid tumors experienced thrombocytopenia,² while the incidence of CIT in hematological tumors reached 75%.³ CIT may lead to chemotherapy dose delay or reduction, and even serious bleeding, which increases the expenditure and worsens the prognosis.⁴

Serum transaminases aspartate transaminase (AST) and alanine transaminase (ALT) are routinely measured to evaluate liver function. ALT is involved in the glucose-alanine cycle, interchange between alanine and pyruvate, and regeneration of glucose consumed by muscle. AST acts in aerobic glycolysis through relocating nicotinamide adenine dinucleotide (NADH) within the mitochondria. These transaminase reactions are essential to the metabolism in muscle and liver cells.⁵ AST/ALT ratio, also termed as De Ritis ratio, has also been proven as an efficient index to diagnose many non-hepatic diseases.⁶ Moreover, it is also an independent prognostic factor for several cancer entities.^7–11 Besides, a recent study has suggested that AST/ALT ratio is inversely associated with palette (PLT) count in severe fever with thrombocytopenia syndrome (SFTS).¹² However, the efficiency of AST/ALT ratio in predicting CIT is unclear.

Cytotoxic chemotherapy can decrease PLT count through several pathophysiological mechanisms. Cytotoxic drugs repress the production of megakaryocytes and PLTs.¹³ In addition, chemotherapeutic drugs can enhance PLT clearance through various immune mechanisms. For example, idiopathic thrombocytopenic purpura (ITP) has been reported in patients receiving fludarabine monotherapy.¹⁴ Currently, treatment options of CIT and related therapeutic benefits are limited. An efficient model based on various variables is needed to predict high-risk CIT groups. Some studies have clarified the relationship between CIT and demographic characteristics, such as the body mass index.^15–17 However, these variables were relatively simple subjective and some show region-related differences.

Therefore, based on objective laboratory indexes, we constructed a model and evaluated its accuracy to predict CIT in a cohort of cancer patients receiving chemotherapy.

Materials and Methods

Patients and Data Collection

This was a retrospective study that included 2043 patients receiving chemotherapy for solid tumor in the Oncology Department of Jinhua Municipal Central Hospital from June 2017 to December 2021. In this study, the primary endpoint was chemotherapy-induced thrombocytopenia predicted by the model. Basic information of the cancer patients, including epidemiological, demographic, clinical, and laboratory data, were obtained from the electronic medical records of the hospital (the Guide Patients Support Care [GPS] information system).

The inclusion criteria were as follows: (1) solid malignant tumors were confirmed by pathological findings; (2) the patient was followed up during all chemotherapy cycles; (3) the patient had blood routinely tested within 1–2 months after the occurrence of CIT; (4) the patient had no other liver diseases, except for malignancies.

The exclusion criterions were as follows: (1) a history of hematologic malignancy confirmed by fluorescence in situ hybridization (FISH) from a bone marrow aspirate and biopsy or peripheral blood review test performed in the prior three months; (2) a history of symptomatic venous thromboembolic events (VTEs) or arterial events; (3) abnormal liver and kidney function confirmed by laboratory indicators within the past three months.

The PLT count test was performed at the beginning of every chemotherapy cycle. A single episode of thrombocytopenia was considered as CIT. The diagnosis of thrombocytopenia was based on the PLT count during chemotherapy. The diagnosis of CIT was made based on the lowest platelet value.

This study complied with the ethical standards of the Institutional Research Council and the Declaration of Helsinki.

Variables Definitions

According to the CTCAE 4.03, thrombocytopenia was divided into four grades: Grade 1, 75×10⁹-99×10⁹/L; Grade 2, 50×10⁹-74×10⁹/L; Grade 3, 25×10⁹-49×10⁹/L; Grade 4, <25×10⁹/L. Anemia was defined as hemoglobin (Hb) concentration <110 g/L. Decreased white blood cell (WBC) count was defined as a WBC count <4000/mL, increased C-reactive protein (CRP) as a CRP level >8 mg/dL, decreased total bilirubin (Tbil) as a Tbil level <25 µmol/L. Hypoalbuminemia was considered as an albumin (Alb) level <30g/L. Increased ALT was defined as an ALT level >50 U/L and increased AST was defined as an AST level >35 U/L. AST/ALT ratio >1 was regarded as abnormal.

Tumors in less than 30 patients were classified as “other tumors”, and those with two or more primary sites were defined as multiple primary cancers (MPC).

Statistical Analysis

Continuous variables were expressed as the mean ± standard deviation (SD) if in normal distribution, or median (interquartile range) if in skewed distribution; categorical variables were described as frequency or percentage. In addition, the Chi-square test (categorical variables) and two-tailed test (continuous variables) were employed to determine differences in means and proportions between groups. Univariate and multivariate logistic regression analyses were applied to identify factors associated with CIT.

A generalized additive mixed model (GAMM) was constructed to explore the relationship between AST, AST/ALT-ratio and PLT. The accuracies of the two variables were compared according to their receiver operating curves (ROCs).

Variables with a p-value less than 0.05 in the multivariate logistic regression analysis were selected. Using them, a nomogram was established to display the predictive accuracy of the final model. The discrimination and calibration of the model were calculated using the area under the ROC curve. A bootstrapping technique was applied using 1000 random data sets (validation set) generated from the original data. All statistical analyses were performed by R software (version 4.1.2) with the “survival”, “survminer”, “rms”, “pROC” and “rmda” packages. P values ≤0.05 were considered statistically significant.

Results

Demographics and Clinical Characteristics of Patients

A total of 2043 patients (1224 [59.91%] males and 819 [40.09%]) females with solid tumors who received chemotherapy during the study period were included. The patients’ detailed demographic and clinical characteristics are summarized in Table 1. Thrombocytopenia occurred in 419 patients, with a frequency of 20.51%. The 5 most common types were (in a downtrend of frequency) cervical (10/27, 37.04%), biliary (13/37, 35.14%), gastric (45/136, 33.09%), ovarian (12/41, 29.27%) and bladder (9/32, 28.12%) cancers.

Table 1 Demographics and Clinical Characteristics of Cancer Patients

The patients’ laboratory results were analyzed. In all patients with CIT, the WBC count decreased in 161 (38.42%) patients. The Hb level dropped below the normal range in 285 (68.02%) patients. The AST level rose above the normal range in 142 (33.89%) patients, and the ALT level in 45 (10.84%) patients. The Tbil level exceeded the upper limit in 57 (13.60%) patients, while the Alb level decreased in 39 (9.31%) patients.

Univariate and Multivariate Analyses Based on Chemotherapy Cycles

A total of 7676 cycles were administered. CIT occurred in 768 (10.01%) cycles (Table 2). The univariate logistic analysis showed that tumor site (P<0.001), AST/ALT ratio (P<0.001), AST (P<0.001), Tbil (P<0.001), Alb (P=0.002), PLT (P<0.001), Hb (P<0.001), WBC (P<0.001), CRP (P=0.020) were associated with CIT of patients with solid tumors (Table 3). In the multivariate logistic analysis, we found that AST/ALT-ratio (P<0.001), AST (P<0.001), Tbil (P=0.04), PLT (P<0.001), WBC (P=0.001) and Hb (P<0.001) were independent predictive factors for CIT.

Table 2 Clinical Characteristics Based on Chemotherapy Cycles

Table 3 Univariate and Multivariate Analysis

Predictive Values of AST/ALT Ratio and AST for CIT

Given that the levels of myeloid, erythroid, and megakaryocytic cells can reflect the hematopoietic function of bone marrow, it has been expected that WBC and Hb are associated with PLT. We further explored whether AST influences the count of PLT. An abnormal AST/ALT ratio was detected after a median of 4 cycles of chemotherapy. The GAMM model showed that both AST/ALT ratio and AST were inversely associated with PLT count (Figures 1 and 2), and the relationship between AST/ALT ratio and PLT showed a stronger linear negative correlation. As shown by the ROC curves, the AUC of AST/ALT ratio was 0.604, while the AUC of AST was 0.569 (Figure 3). Obviously, AST/ALT ratio was more accurate than AST in predicting CIT (P=0.045).

Figure 1 Shift trend of AST and PLT of cancer patients analyzed by GAMM.

Figure 2 Shift trend of AST/ALT-ratio and PLT of cancer patients analyzed by GAMM.

Figure 3 The compared ROC between AST/ALT-ratio and AST.

Predictive Value of Nomogram for CIT

In the multivariate logistic regression model, AST/ALT ratio, AST, PLT, WBC and Hb were statistically significant predictors for CIT. Therefore, a nomogram containing AST/ALT-ratio, Tbil, Alb, PLT, WBC and Hb was established to predict CIT of chemotherapy-treated patients (Figure 4). The AUC value of the model was 0.793 (sensitivity 0.543, specificity 0.930), which indicated that the prediction model has a high accuracy (Figure 5). In addition, the calibration plots revealed a good match between the CIT frequencies predicted by the nomogram model and observed in the real setting (Figure 6).

Figure 4 The nomogram model for CIT probability prediction based on laboratory indicators. Abbreviations: Tbil, total bilirubin; Alb, albumin; PLT, platelet; WBC, white blood cell; Hb, hemoglobin; AST/ALT-ratio, aspartate aminotransferase/alanine transaminase ratio; Risk; Risk probability of developing thrombocytopenia.

Figure 5 The area under the ROC curve (AUC). AUC:0.793, Sensitivity: 0.543, Specificity: 0.930.

Figure 6 The calibration curve for predicting CIT in the validation set: the ideal prediction curve has a good fit with the actual observations.

Discussion

In this study, we found that baseline laboratory parameters, including PLT, WBC, Hb, AST, Tbil and AST/ALT-ratio, were significantly associated with CIT risk. There was an inversely relationship between AST/ALT ratio and PLT count, as well as between AST and PLT count. Besides, a nomogram incorporating the six significant parameters demonstrated a strong predictive ability.

In our cohort, the incidence of CIT during the chemotherapy course was 20.51%, higher than those in previous studies.^18,19 The incidences in different cancer types were slightly different from other investigations.^20–24

To the best of our knowledge, our study was the first to evaluate the potential predictive value of AST/ALT ratio and AST for CIT. Furthermore, we found AST/ALT-ratio and AST were both inversely associated with PLT. In a prognostic study of severe fever with SFTS, an evident inverse correlation was found between AST/ALT ratio and PLT count.¹² In other studies of SFTS, some inflammatory factors were up-regulated in infected HepG2 cells. Proinflammatory cytokines, including IL-6 and tumor necrosis factor-alpha (TNF-α), increase dramatically with disease severity.²⁵ Patients who received allogeneic hematopoietic stem cell transplantation exhibited high levels of TNF-α, interferon-gamma (IFN-γ), and IL-2, as well as ALT and AST.²⁶ These indicate that the immune system is disrupted in liver injury. Previous studies have highlighted that PLTs are essential to cellular inflammation and can be used to score hepatic fibrosis. Xu et al have found that a low PLT count is associated with cirrhosis and progression to hepatic decompensation among patients with chronic hepatitis C.²⁷ In patients receiving chemotherapy, the AST/ALT ratio increases and the PLT count decreases, indicating the damage to hepatic function. Chemotherapy-induced liver injury (CALI) is also a common adverse event in cancer patients. The evolution of CALI is usually accompanied by elevations of ALT, AST, and Tbil. Idiosyncratic CALI has been explained as a result from adaptive immune attack in previous reports, with immune cell-mediated liver injury playing a critical role.²⁸

Some molecules trigger the adaptive immune attack on the liver by up-regulating costimulatory factors on professional antigen-presenting cells and promoting the release of cytokines and chemokines. In addition, some cytokines, such as TNF-α and IFN-γ, can make hepatocytes more susceptible to drug-induced stress by shifting cellular responses away from cell survival towards cell death.²⁹ Considering that CIT involves specific immune responses induced by chemotherapeutic drugs, we propose that the link between elevated AST and CIT may be mediated by immune-related mechanisms. Currently, the immune-related mechanisms underlying CIT have not been fully clarified. No immunosuppressive agents have been used in the routine treatment of CIT. However, in some cases whose thrombocytopenia is known to be caused by immune mechanisms, such as ITP caused by fluadabine, corticosteroids or gamma globulin can be used to treat thrombocytopenia, and the therapeutic effect is significant.³⁰

Interestingly, serum AST/ALT ratio has been found related to a high mortality rate in previous cancer models. In patients with oral and oropharyngeal cancer, AST/ALT ratio >1.44 is an independent prognostic factor for poor cancer-specific survival and overall survival (OS).³¹ Furthermore, AST/ALT ratio has been validated as a prognostic marker for disease-free survival (DFS) in stage II–III non-cancer patients.³² Besides, it can also predict the prognosis of gastric, prostate and advanced-stage pancreatic cancer.^33–35 In the current study, we addressed that the AST/ALT ratio tightly correlates with CIT risk. However, the mechanism contributing to the increase in AST/ALT ratio and the decrease in PLT after chemotherapy is still unclear, and should be uncovered by future studies.

Severe thrombocytopenia may lead to bleeding, which require platelet transfusion. Few models have been developed to predict CIT risk based on clinical and laboratory variables.³⁶ These predictive models only incorporate specific predictors and their application is limited to a certain cancer type. In the present study, our model was based on the data from a whole spectrum of cancers, thus largely expanding its clinical applicability. Compared with previous models, our model absorbed real-time laboratory indicators measured in those who had passed the myelosuppression phase but not received the next cycle of chemotherapy. These indicators, easily measurable, can efficiently help to distinguish the patients at a high risk of CIT.

However, several limitations should be mentioned. First, our study lacked external data validation. Second, our study was a retrospective and single-center study, and therefore, the potential biases could not be avoided. Such drawbacks will be addressed in our future studies.

In conclusion, our model was accurate in predicting CIT risk, and AST/ALT ratio was closely associated with PLT count in patients with solid tumors. Our model may provide a possibility to design individualized treatment for cancer patients.