One aim of the present study was to evaluate the impact of pretreatment values, age, sex, CCRT, cumulative radiation dose, and location of irradiation on CBC parameters during radiotherapy. Multiple linear regression analyses were performed. The model explained approximately 70% of the variance in Hb level during radiotherapy. Statistically significant variables included baseline Hb level; age; male sex; CCRT; and pelvic, brain, bone, and head and neck sites. Only cumulative radiation dose and a thoracic site were nonsignificant. During the therapy, regression analysis explained approximately 10%, 30%, 6%, and 40% of the variance in WBC count, neutrophil level, ANC, and platelet count, respectively.

In correlation analysis, despite the weak correlation between cumulative dose and CBC parameters, negative associations of WBC count, ANC, and particularly platelet count were evident. In contrast, the strength of the association of Hb level was remarkably weak. Besides, in multiple regression analysis, the coefficient of cumulative dose was not statistically significant for Hb level. These results are consistent with the results of a study by Sini and colleagues, which showed that the Hb level remained stable throughout postoperative whole-pelvis intensity-modulated radiotherapy (IMRT) in prostate cancer patients, and thereafter for at least 1 year, while WBC, ANC, and platelet nadirs occurred at the treatment midpoint, with a similar magnitude of decline, and remained below baseline thereafter [5]. These results can be explained by the lifespan of each blood cell. In agreement with Casarett’s classification, hematopoietic stem cells are the most radiation sensitive of cells (class 1; vegetative intermitotic cells), while polymorphonuclear granulocytes and erythrocytes are the most radiation resistant of cells (class 5; fixed postmitotic cells) [6]. Therefore, the concentration of each blood cell type depends mainly on their lifespan. Red blood cells (RBC) have a lifespan of approximately 120 days, whereas neutrophils and platelets have lifespans of about 5.4 days and 9–12 days, respectively. Compared to neutrophils and platelets, RBC can live long enough for bone marrow at unirradiated areas to be able to compensate for decreased RBC production or for bone marrow at the irradiated area to regenerate itself, while the others cannot. This is why the Hb level was not associated with cumulative dose, whereas WBC count, ANC, and platelet count were negatively associated with the variable.

On the basis of dosimetric parameters, the Hb level was reported to be affected by a specific radiation dose delivered to a defined volume. In a systematic review by Corbeau et al. on the correlation between pelvic bone marrow radiation dose and hematologic toxicity in cervical cancer treated with CCRT, only V10 and V20 were found to be correlated with Hb nadirs [7]. Whole iliac bone V20 > 90% was revealed by Lewis et al. to be predictive of grade ≥ 2 anemia in cervical cancer patients receiving postoperative CCRT with IMRT [8]. Among patients with cervical cancer treated with postoperative CCRT using IMRT, whole-pelvis bone marrow (WP) V30, V40, V45, and V50; iliac bone marrow (IL) V20, V40, V45, and V50; and lower pelvic bone marrow (LP) V30, V40, V45, and V50 were associated with anemia in a correlation analysis presented by Yang et al. In multiple regression analysis, the side effect was negatively correlated with WP V30, IL V40, and LP V40 [9]. Regarding the data of cervical cancer patients treated with CCRT and IMRT published by Chen and colleagues, Hb nadirs were positively correlated with baseline values and negatively correlated with relative lower pelvis V10, V25, V50, and Dmean. The cutoff values of these variables were 96.150 g/L, 74.62%, 44.290%, 7.258%, and 2020.850 Gy, respectively [10]. These findings conflict with the results of the present study. All the mentioned studies demonstrate that a sufficient dose delivered to an adequate volume could lead to significant Hb level reduction. However, in the previously mentioned systematic review, dosimetric parameters including V10, V20, V30, and Dmean were correlated with WBC nadirs. The results were consistent with ANC nadirs. Additionally, the baseline values of each parameter were shown to be predictive for their nadir values. These relationships were comparable to the present study. Nevertheless, no dose–volume parameter was reported to influence the platelet count in the systematic review [7].

There was a significant imbalance between the development and validation cohorts due to the use of random sampling in the present study. However, this imbalance could help in examining the generalizability of the models to populations that differ substantially from the development cohort. To evaluate whether the model is valid, performance metrics should be considered. With acceptable values in both development and internal validation datasets, the next step should involve external validation. In the present study, it could be agreed that the performance metrics for Hb level were admissible, with fairly high adjusted R‑squared and low mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE). Compared to the other parameters with adjusted R‑squared less than 50%, as well as large MAE, MSE, and RMSE, caution should be exercised during interpretation of the results. To improve the performance of the models, other regression analyses should be used to augment R‑squared and minimize the errors. In addition, machine learning (ML) has been shown to precisely predict outcomes in both medical and nonmedical scenarios in a vast number of publications and media [11, 12]. Commencing ML for CBC parameter prediction could represent a huge step in the development of patient care for radiotherapy, accurately identifying patients who require CBC monitoring during radiation treatment, which could result in better quality of life for patients and avoid superfluous expense.

A more pronounced incidence of grade ≥ 3 neutropenia in the elderly (≥ 70 years) was reported in the secondary analysis of the RTOG 94–10 trial comparing CCRT with different fractionation schemes to sequential chemoradiotherapy in patients with non-small cell lung cancer (NSCLC) [13]. The result was consistent with the secondary analysis of the NCCTG 94–24–52 trial comparing different CCRT fractionations in NSCLC patients. The authors revealed that elderly patients (≥ 70 years) had more grade ≥ 4 hematologic toxicity and leukopenia than younger ones [14]. On the contrary, in a nomogram established from the data of 3786 patients receiving radiotherapy alone by Takeda and colleagues, for the population of age 70 years and older, the older the patient gets, the less neutropenia occurs; however, this association was not found for anemia [15]. The results of the previous studies were inconsistent with those of the present study. Only Hb level and platelet count were significantly negatively correlated with age, and the associations were very weak. Notably, there were only 66 courses with patients aged at least 70 years (13.3%) in the present study. Given the small elderly population, the results are unlikely to be comparable to those of other studies. Nevertheless, there was no association found between hematological complications and age in the systematic review by Corbeau et al. [7]. This emphasizes the need for further research to investigate the association between patient age and hematological toxicity with exceptionally robust methodology.

Due to being a retrospective study, missing data can be expected, and neither randomization nor blinding was performed. Hence, the results should be interpreted with caution. On top of that, since all the radiation treatments were conducted using 3DCRT, it is questionable whether the results can be generalized to IMRT, especially in terms of cumulative dose, because bone marrow tends to receive different doses between these techniques. Besides, external validation with different datasets is needed before implementation of the model to evaluate its reproducibility and particularly generalizability, making sure that the model is transportable to different populations.

In conclusion, CBC parameters during radiotherapy were positively correlated with their baseline values. For Hb level, the model could explain the variance fairly; nonetheless, it was not significantly associated with the received cumulative dose. Although external validation is essential before implementation of the model is merited, the results could be used as a guide to individualize patient care before and during radiation treatment, especially for specifying patients who might require special attention or benefit from prophylactic management.