criterion.

Beta-thalassemia major is a highly severe disorder and represents the most common hereditary anemia in Iran. The disease is characterized by early clinical manifestations, which typically appear between three and six months of age. To prevent mortality and ensure patient survival, regular blood transfusions administered at scheduled intervals are essential [1]. This condition is associated with physical deformities, skeletal complications, weakness, and growth retardation, all of which have significant psychological and social implications in addition to their physical consequences [2], [3]. Fortunately, recent advances in medical science have enabled timely treatment and continuous care for affected patients, and the disease is no longer considered as fatal as it once was. In Iran, the major prevalence clusters of thalassemia are located in the coastal and neighboring provinces, including Khuzestan, Bushehr, Hormozgan, Sistan and Baluchestan, Gilan, and Mazandaran [4].

Today, statisticians employ appropriate statistical models to establish relationships between a set of explanatory variables and a response variable. These models enable researchers to explain and predict the response variable based on the explanatory variables [5]. In survival models, also known as time-to-event or failure-time models, the response variable represents the time until the occurrence of a particular event. Accordingly, the identification of causes and determinants of mortality among patients with beta-thalassemia major can be effectively investigated using survival analysis models. One of the most widely used and popular approaches in survival analysis is the semi-parametric Cox proportional hazards model, primarily because it does not require specification of the probability distribution of survival times. Another approach involves the use of parametric survival models, such as the Weibull, log-normal, log-logistic, and Gompertz models. Among these, the Weibull model is considered more general and is more frequently applied than other parametric models because its hazard function is not constant over time. Furthermore, the inclusion of an additional parameter, referred to as the shape parameter, provides greater flexibility to the model. In survival studies, patients with similar risk factors may nevertheless exhibit different survival durations. This variation may be attributable to the influence of one or more unobserved variables not included in the model. In such situations, semi-parametric or parametric frailty models can be employed. In these models, a significant frailty component indicates that individuals with identical explanatory variables and similar characteristics do not necessarily experience the same survival times, suggesting that one or more additional unmeasured covariates may be required to improve model fit. To address this issue, a random effect component is incorporated into the model, and the significance of this component can be used to evaluate the heterogeneous effects of risk factors across patients [5], [6]. Most previous studies in this field have primarily adopted semi-parametric approaches. Therefore, conducting the present study using parametric methods, in addition to identifying mortality risk factors, may contribute to the introduction of a more efficient and better-fitting model. The aim of this study was to fit various parametric survival models alongside the semi-parametric Cox model and to determine the most appropriate model for patients with beta-thalassemia major.

In the present study, data from 190 patients with beta-thalassemia major who had attended a treatment center in Isfahan between 2011 and 2021 were analyzed. The explanatory variables included patient age, age at disease onset, age at first Desferal administration, age at initiation of blood transfusion, splenectomy status (splenectomized vs. non-splenectomized), sex, blood transfusion status (regular, irregular, or no transfusion), parental relationship (unrelated, distant relatives, or close relatives), type of Desferal administration (subcutaneous, pump infusion, both methods, or none), maternal educational level (illiterate, below diploma, diploma, or university education), patient age group (born before 1986 vs. born after 1986), and place of birth (coastal vs. non-coastal regions). The response variable was defined as the duration between disease diagnosis and either death or continuation of treatment. For the survival analysis of patients, parametric models including the exponential, Weibull, log-normal, log-logistic, and Gompertz models, as well as the semi-parametric Cox proportional hazards model, were employed. In addition, the significance of the frailty component was assessed for all fitted models. To evaluate the effects of explanatory variables and to identify the optimal model, variables found to be significant in the univariate analyses were initially entered into the corresponding multivariable models. Subsequently, variables with borderline significance were added sequentially according to ascending p-values. If the inclusion of a variable resulted in a higher Akaike Information Criterion (AIC) value compared with the previous step, that variable was removed from the model. In other words, model selection was based on the AIC, whereby smaller AIC values indicate a more appropriate and better-fitting model [7].

In this study, data from 190 patients with beta-thalassemia major who had been referred to a treatment center in Isfahan between 2011 and 2021 were collected. The mean age of the patients was 30.71 years, with a standard deviation of 0.52. Among the study participants, 99 (52%) were female. Regarding place of residence, 92 individuals (48%) lived in coastal areas. In terms of maternal education, 59 cases (31%) had mothers with below-diploma education or were illiterate, while this figure for fathers was 58 (30%). A total of 139 patients (73%) were born in or after 1986. Splenectomy had been performed in 80 patients (42%). Regular blood transfusion was reported in 141 cases (74%). Regarding Desferal administration, 61 patients (32%) received subcutaneous treatment. Additionally, 70 patients (37%) had no family relationship or their degree of consanguinity was unknown. The mean survival time estimated using the Kaplan–Meier curve was 40.65 years. Moreover, the mean survival times for males and females were 40.12 and 39.48 years, respectively. The distribution of patients’ characteristics is presented in Table 1.

The AIC was used to select the best-fitting model. The model with the smallest AIC value was considered the most appropriate model. The AIC values for each model are presented in the table below.

Among the frailty models, only the Weibull model achieved convergence, whereas the remaining frailty-based models did not converge; therefore, the AIC could not be computed for those models. According to the results presented in Table 3, the parametric log-normal model exhibited the lowest AIC value and was thus selected as the best-fitting model. Consequently, a multivariable survival analysis of patients with beta-thalassemia major was conducted using this model, and the statistically significant variables were identified. The findings indicate that patients’ place of birth, age, maternal education level, age at first Desferal administration, and age at initiation of blood transfusion were statistically significant. For instance, the significance of the place of birth suggests that, based on the selected model, patients living in coastal provinces had lower survival compared with other patients.

Beta-thalassemia major is the most prevalent form of anemia in Iran and other parts of the world, and various survival analysis studies have been conducted on patients affected by this disease. The present study aimed to fit standard parametric models, the semi-parametric Cox model, and frailty models in order to identify the most appropriate model for survival analysis of these patients. Ultimately, the log-normal parametric model was selected as the best-fitting model. In a similar study, Borgna-Pignatti et al. [8] concluded that the survival rate of individuals born after 1970 was higher than that of those born before 1970. Rodbari et al. [9] demonstrated that regular blood transfusion, attention to blood screening, and increased family awareness were effective in preventing complications and improving the life expectancy of patients with thalassemia. Ladis et al. [10] estimated the survival probability of 647 patients with beta-thalassemia major using the Kaplan–Meier method and compared the results using the log-rank test. They showed that survival was higher among individuals born after 1975 compared with those born earlier. Latifi and Zandian [1] investigated the survival of patients with beta-thalassemia major in Ahvaz using life tables, Kaplan–Meier curves, and the log-rank test. Their results indicated that ethnicity (Fars, Arab, and Lur) and gender had no significant effect on survival. Yavarian et al. [11] reported the survival of 101 patients with thalassemia using the Kaplan–Meier curve, showing that 68% of patients survived up to the age of 20 years, and only half of the patients were expected to survive to 30 years of age. The findings of the present study indicated that place of birth, age at disease onset, age at first Desferal administration, maternal education level, and age at initiation of blood transfusion had a statistically significant effect on patient survival. Unlike previous studies, birth cohort (year of birth) did not show a significant effect on survival in this study; however, results related to factors such as age at disease onset, maternal education, and age at initiation of blood transfusion were consistent with previous findings. In earlier studies, survival analysis was mainly conducted using the Kaplan–Meier method, log-rank test, and univariate models, while parametric and frailty models were not applied; therefore, direct comparison with these approaches was not possible. One of the limitations of this study was the presence of a considerable amount of missing data among explanatory variables, which led to a reduction in the effective sample size during model fitting. Given that the log-normal parametric model was identified as the best model for survival analysis of beta-thalassemia major in this study, its application is recommended for future research in this field. Furthermore, increasing parental awareness, particularly maternal education, timely blood transfusion, and proper blood screening for early diagnosis may contribute to improved survival outcomes in these patients.

All aspects of the research and manuscript preparation were carried out by the author. The author has read and approved the final version of the manuscript.

This study did not receive any specific funding from public, commercial, or non-profit funding agencies.

All data are included in the text.

The author declares that he does not have any conflict of interest.

The author has given consent for the publication of this manuscript.

This study does not involve any research conducted on human participants or animals.