The prediction of disease chronicity when immune thrombocytopenia is first diagnosed Authors: Anselm Chi-Wai Lee Published: 2026 Introduction Immune thrombocytopenia (ITP) is the most common bleeding disorder in children, affecting 2 to 5 individuals under the age of 18 years per 100,000 annually (1). The immune mediated wastage and impaired production of platelets can be broadly classified into primary and secondary disorders. Secondary ITP refers to the occurrence of isolated thrombocytopenia in the presence of an underlying chronic illness such as other autoimmune diseases or immunodeficiency states (2). Otherwise, the condition is categorized as primary ITP. ITP that happens along with viral illnesses or after immunizations behaves similarly to primary ITP and is therefore considered under such category. Chronic ITP–the Changing Landscape and Treatment Prior to 2009, chronic ITP was defined as persistent thrombocytopenia (platelet <150×109/L) beyond six months after the initial diagnosis. Based on the results of 12 cases series, a review in 1996 reckoned that 390 (24.4%) out of 1,597 children diagnosed with ITP subsequently lapsed into chronicity (3). After 2009, the revised definition of disease duration more than 12 months, with platelet <100×109/L, as proposed by the International Working Group has been universally adopted (2). Despite a longer duration after initial diagnosis and a stricter threshold for platelet count (and thus fewer cases of chronic disease expected), studies from single centers published after 2010 reported similar rates of chronic ITP (Table 1). A probable explanation is that earlier studies of childhood ITP might have included transient thrombocytopenic illnesses with platelet counts 100–149×109/L that were not genuinely immune-mediated in nature. Another probable explanation is that there are higher attrition rates with dropping out of less serious cases over the longer duration of observation (18). This may explain the higher rate of chronic ITP observed from voluntary reports from a multicenter registry (4). Table 1The published rates of development of chronic ITP after 2010.ReferencesNumber of patientsAge, yearChronic ITPRisk factorsRemarksSingle-center studies Evim, 2014 (4)201Median, 5.8; Range, 0.6–17.360/199 (30.2%)- Female sex2 splenectomized cases excluded- Age >10 year Jung, 2016 (5)64Median, 1.0; Range, 0.1–15.09/64 (14.1%)- Older age- Absence of infection- Insidious onset Kalfon, 2018 (6)62Mean, 6.1; S.D., 4.218/62 (29.0%)- Older age- Multiple types of skin bleeding Gungor, 2019 (7)211Mean, 5.4; S.D., 4.160/211 (28.4%)- Female sex- Older age- Absence of infection- Higher platelet Sun, 2020 (8)234Range, 0.1–13.034/234 (14.5%)- Age between 1 and 7 year- Bleeding onset to diagnosis >1 month- ALC <4.905×109/L Chan, 2022 (9)111Median, 3.226/111 (23.4%)- Older age- Higher platelet- Insidious onset Alpdogan, 2023 (10)123Mean, 6.4; S.D., 4.030/123 (24.4%)- Older age- Female sex- Insidious onset Cheng, 2023 (11)497Median, 3.8; IQR, 1.4–7.6104/497 (20.9%)- Older ageOnly platelet <30×109/L included- Insidious onset Kulmala, 2025 (12)403Mean, 6.5; S.D., 4.895/367 (25.9%)- Older age- Female sex- Insidious onset- Higher platelet- Absence of infection Lee, 2025 (13)Total, 52-10/52 (19.2%)Only aged 1–48 months included1–12 mo, 221/22 (4.5%)13–48 mo, 309/30 (30%) Alali, 2026 (14)47Median, 5.0; IQR, 5.0–9.012/47 (25.5%)- Transient response to treatment Pregeljc, 2026 (15)271Median, 4.0; IQR, 2.0–10.572/240 (30.0%)- Older age31 cases excluded when alternative diagnoses found- Absence of infection- Higher plateletTotal, 530/2,207 (24.0%)Multicenter studies Grimaldi-Bensouda, 2017 (16)257Mean, 6.8; S.D., 5.151/211 (24.2%)- Older age- Female sex- Higher platelet Schifferli, 2018 (17)1,639Mean, 5.7; S.D. 4.4479/1,639 (29.2%) Heitink-Polle, 2018 (18)200 Only platelet ≤20×109/L includedIVIg group, 100Median, 3.6; Range, 0.3–16.110/100 (10.0%)Control group, 100Median, 4.5; Range, 0.5–16.612/100 (12.0%)ALC, absolute lymphocyte count; ITP, immune thrombocytopenia; IQR, inter-quartile range; IVIg, intravenous immunoglobulin; mo, months old; S.D., standard deviation. The published rates of development of chronic ITP after 2010. ALC, absolute lymphocyte count; ITP, immune thrombocytopenia; IQR, inter-quartile range; IVIg, intravenous immunoglobulin; mo, months old; S.D., standard deviation. The immediate management of bleeding in children with chronic ITP follows the same principles as in acute ITP (19). Strategically, however, the use of steroid-sparing, potentially disease-modifying and experimental treatment is open for consideration. Splenectomy, with sustained complete remission rates up to 76% (20), used to be the only option in the last century. To date, thrombopoietin receptor agonists and mycophenolate mofetil are often used as steroid-sparing treatments when standard first-line therapies with corticosteroid and/or intravenous immunoglobulin (IVIg) fail to induce sustained remission (21). In selected patients, the use of rituximab as B-cell depleting treatment results in long-term remission in 30% of children (22). Will Predicted Chronicity Affect the Management of Acute ITP? Various clinical and laboratory features present during the early weeks of diagnosis of ITP have been found to be associated with subsequent disease chronicity. An older age, the female sex, an insidious onset, and a higher platelet count are classical risk factors for the development of chronic ITP (3). Nonetheless, the current American and international guidelines do not take into consideration the propensity to chronicity when recommending treatment during the first 12 months of diagnosis of ITP (21,23). In their recent study, Hillier et al. argued that, by providing an accurate prediction of the odds for developing chronic ITP, the medical team may be able to change the course of the disease, identify secondary causes for the immune-mediated thrombocytopenia, and thus at a better position on counseling the affected family (24). They developed a multivariate logistic regression model, incorporating the patients’ age, sex, presenting platelet and absolute lymphocyte counts, baseline immunoglobulin (Ig) G, IgA, and IgM levels, direct antiglobulin test (DAT) positivity, and known secondary causes at diagnosis to predict the likelihood of subsequent chronic ITP. The study started with 611 children from two hospitals as the training cohort, and corroborated the statistical model with a validation cohort of 161 cases from another two institutions. Their model is available at https://opal.shinyapps.io/citp-rm/. The overall performance of the predictive model is considered adequate based on an acceptable [area under the receiver operating characteristic curve (AUROC) of 0.774, (with usual values between 0.5 and 1.0; 0.5 represents random guess and 1.0 indicates perfect prediction with 100% accuracy)], and a moderate discriminatory power between predicted probabilities and observed outcomes (Somers’ Dxy of 0.55, with rank correlation coefficient values between −1 and 1; 0 represents random guess and 1 indicates perfect discrimination). Limitations of the Prediction Model As the authors have alluded to, the validity of the model is limited by the retrospective nature of the data. When the Intercontinental Cooperative ITP Study Group first analyzed the outcomes of children registered retrospectively (25), treatment with IVIg was associated with a higher chance of normal platelet counts (≥150 × 109/L) at six months after diagnosis (odds ratio 1.81; 95% confidence interval, 1.25–2.64). The results were statistically significant by matched pair analysis and were independent of sex, age, country of origin, platelet count at diagnosis, or preceding infections. Their findings suggested that treatment of acute ITP with IVIg was predictive of a lower risk of chronic ITP. A subsequent randomized controlled study, however, did not find this to be true (18). It should be noted that the patient cohorts recruited for training and validation purposes might not be representative of childhood ITP patients in general. The rate of chronic ITP (201/611 or 32.9%) from the training cohort was higher than those reported from the literature (Table 1). The rates were 32.0% (176/550) and 41.0% (25/61) when calculated separately from the two contributing institutions, respectively. This suggests that the selected data for training might have originated from a biased patient population. Similarly, the proportion of chronic ITP from the validation cohort was 37.3% (60/161), with rates of 45.8% (38/83) and 28.2% (22/78) when calculated separately for the individual hospitals, respectively. Another source of potential bias could be attributed to the missing data which seemed to happen more often with the training data from the institution with an unusually high rate of chronic ITP. Therefore, for these concerns, further validation with prospective studies is required. Will Risk Prediction Alter the Course of ITP? The proposition that the propensity to develop chronic ITP can be hampered with treatment is appealing but currently there are no licensed medications to fulfill this purpose. As stated before, prior claims that treatment with IVIg might reduce the chance of developing chronic ITP cannot be substantiated. Neither corticosteroids nor thrombopoietin receptor agonists such as eltrombopag or romiplostim alter the natural history of childhood ITP. The off-label use of rituximab targeting CD19-positive B cells may be associated with long-term disease remission in up to one-third of recipients with chronic ITP, but its use during the first 12 months of diagnosis has not been tested and hence cannot be recommended. Similarly, treatment targeting CD38-positive B cells may hold similar promises but the evidence is currently too limited for general recommendation (26). Can Risk Prediction be Useful for Counseling? Although the prediction model developed by Hillier et al. appears to be statistically robust, its impact in the real world has not been tested. Four archival cases from the author’s record with sufficient data to test the prediction model are tabulated in Table 2. If the predicted risk is arbitrarily categorized into low-risk (predicted risk <33.3%), high-risk (predicted risk >66.7%), and intermediate-risk (predicted risk 33.3% to 66.7%), three of the four cases carry intermediate-risk of developing chronic ITP, which is not very different from the odds of tossing a coin. This limited set of patients should not be viewed as contradicting evidence against Hillier’s prediction model, but it illustrates the frustrations in pediatric hematology where disease evolution can rarely be accurately foretold. Table 2Predicted risk of chronic ITP from four anonymized cases of the author’s records according to the prediction model described by Hillier et al. (24)CasesSexAge groupPredicted risk of chronic ITP,mean (95% confidence interval) (%)Final status1MaleToddler8.9 (5.7–13.6)Chronic ITP requiring regular treatment2MaleSchool-age62.1 (26.6–88.2)Persistent ITP3MaleSchool-age37.4 (24.5–52.3)Chronic ITP not requiring treatment4FemaleAdolescence54.8 (32.3–75.4)Chronic ITP requiring regular treatmentToddler, aged 1–5 years old; School-age, aged 6–11 years old; Adolescence, 12–18 years old. ITP, immune thrombocytopenia. Predicted risk of chronic ITP from four anonymized cases of the author’s records according to the prediction model described by Hillier et al. (24) Toddler, aged 1–5 years old; School-age, aged 6–11 years old; Adolescence, 12–18 years old. ITP, immune thrombocytopenia. The prediction model also does not take into consideration the heterogeneous and dynamic nature of chronic ITP in children. A significant proportion of children with chronic ITP do not have severe thrombocytopenia and most of their bleeding episodes, if any, are classified as mild bleeding (27,28). Thrombocytopenia continues to improve with time and up to 50% of them will eventually remit (29,30). Clinicians should not rely on the prediction model as the only tool when counseling parents on childhood chronic ITP. Other Prediction Models Hillier et al. (24) is not the only group that attempted to predict platelet recovery in childhood ITP. Schmidt et al. (31), using data from the Nordic countries (377 patients) as the training cohort and those from the Netherlands (194 patients) as the validation cohort, developed the Childhood ITP Recovery Score to predict early recovery. Their prediction model was primarily built to foretell platelet recovery at three months and thus distinguishing acute from persistent and chronic ITP. However, the final data could also differentiate complete recovery rate at 12 months of follow-up, with an acceptable model performance similar to the current study (AUROC of 0.70 for the exploratory cohort and 0.71 for the validation cohort). However, Schmidt et al. only included patients with initial platelet counts ≤20×109/L and hence their model is not applicable for all children with ITP. Others experimented with machine learning models based on traditional prognostic factors similar to those employed by Hillier et al. Kasser et al. (32) analyzed a longitudinal cohort of 944 children from the Intercontinental Cooperative ITP Study Group using random forest prediction models. Their models identified age, sex, cutaneous bleeding, leukocyte, platelet, and hepatitis C at diagnosis as pivotal predictors. The model performance was not strong (AUROC of 0.693) and two patients had to be flagged for chronic ITP in order to identify one true case. Zhang et al. from China (33) recruited 156 subjects. Their extreme random trees model (AUROC of 0.864) outperformed the random forest and other models. Age, insidious onset, immunoglobulins A and G, and anti-platelet antibodies emerged as potent predictors. Lejeune et al. (34), enlisted 161 children from five French hospitals and successfully developed a model based on platelet recovery kinetics. They predicted chronicity using platelet counts from 15 or 30 days after the diagnosis of ITP (AUROC of 0.82). Although both models from Zhang et al. and Lejeune et al. seem to perform reasonably well, they are based on small numbers for the training and validation processes. The remarkably good performance of Zhang et al.’s model may also be related to their cohort heavily represented by patients with chronic ITP (43.6%). It should be pointed out that each of these studies has different patient inclusion criteria and hence direct comparison among the models may not be straightforward. Concluding Remarks The work by Hillier et al. in providing a statistically robust prediction model for the odds of developing chronic ITP in pediatric patients should be applauded. However, its applicability in the real world should be tested prospectively and refined according to the observations generated. It would be interesting if the different models reported recently can be compared or integrated into a more robust platform. Supplementary The article’s supplementary files as Acknowledgments None. Ethical Statement: The author is accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Footnotes Provenance and Peer Review: This article was commissioned by the editorial office, Translational Pediatrics. The article has undergone external peer review. Funding: None. Conflicts of Interest: The author has completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0739/coif). The author has no conflicts of interest to declare. References Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0739/prf Supplementary Materials The article’s supplementary files as