Machine-learning Based Prediction Model in Primary Immune Thrombocytopenia

• Conditions: Immune Thrombocytopenia; ITP

• Registration Number: NCT05116423
• Lead Sponsor: Peking University People's Hospital

Brief Summary

This study developed the first prediction model for risk of critical ITP bleeds for ITP inpatients using a novel machine learning algorithm. This model has been implemented as a web-based model so that clinicians can obtain the estimated probability of critical ITP bleeds for ITP inpatients. The objective of this study is to prospectively and externally validate the risk of critical ITP bleeds in newly admitted ITP patients.

Detailed Description

Primary immune thrombocytopenia (ITP) is a common acquired autoimmune disease characterized by reduced platelet production and increased platelet destruction due to autoimmune disorders, as patients present with low platelet counts and a high risk of bleeding. Although most ITP patients present a good prognosis, the rare but important critical ITP bleeds events are the threatening-life complication to ITP patients, severely affecting their prognosis, quality of life and treatment decisions.

More recently, the development of clinical prediction models has provided powerful tools for precision diagnosis and early intervention of diseases, especially the application of machine learning methods. Machine learning approaches can overcome some of the limitations of current risk prediction analysis methods by applying computer algorithms to large data sets with numerous multidimensional variables, capturing the high-dimensional nonlinear relationships between clinical features to produce data, drive outcome prediction.

It suggests an unmet need for personalized patient management strategies and an urgent need for effective tools to predict the risk of critical ITP bleeds in hospitalized patients in medical practice.

Here, we aim to integrate clinical and laboratory data based on a nationwide multicenter study in China to build a clinical prediction model. In particular, we also perform external and prospective validation with large sample sizes to improve the robustness and utility of our models.

It is a simple and convenient tool to quickly assess newly admitted ITP patients and achieve early identification and intervention for those at high risk of life-threatening bleeding events, thus reducing disability and mortality rates in the future.

Study Timeline

• Study First Submitted: 2021-11-09
• Study First Submitted QC: 2021-11-09
• Study Start: 2021-11-10
• Study First Posted: 2021-11-11
• Status Verified: 2022-01
• Last Update Submitted: 2022-01-24
• Last Update Posted: 2022-02-08
• Primary Completion: 2022-03-01
• Study Completion: 2022-06-30

Recruitment & Eligibility

• Status: UNKNOWN
• Sex: All
• Target Recruitment: 100

Inclusion Criteria

1. Confirmed ITP diagnosis;

Exclusion Criteria

1. Received chemotherapy or anticoagulants or other drugs affecting the platelet counts within 6 months before the screening visit;
2. Current HIV infection or hepatitis B virus or hepatitis C virus infections;
3. Maligancy;
4. Female patients who are nursing or pregnant, who may be pregnant, or who contemplate pregnancy during the study period; a history of clinically significant adverse reactions to previous corticosteroid therapy
5. Have a known diagnosis of other autoimmune diseases, established in the medical history and laboratory findings with positive results for the determination of antinuclear antibodies, anti-cardiolipin antibodies, lupus anticoagulant or direct Coombs test;
6. Patients who are deemed unsuitable for the study by the investigator.

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Primary Outcome Measures

Name	Time	Method
Performance of model	3 months	Area under receiver operating characteristic curve (AUC) of the model in predicting critical ITP bleeds in patients with ITP.

Secondary Outcome Measures

Name	Time	Method
Comparison between different machine learning algorithms used in the model	3 months	Comparison of sensitivity and specificity of different machine learning algorithms used in the model.

Trial Locations

Locations (1)

Peking University Insititute of Hematology, Peking University People's Hospital; 🇨🇳 Beijing, Beijing, China