Preterm neonates often receive platelet transfusions when their platelet count is low to prevent bleeding. However, it is currently unclear which infants benefit from such transfusions. A recent randomized controlled trial (PlaNeT-2/MATISSE trial) showed that the higher platelet count threshold for transfusion was associated with a higher risk of major bleeding or death. Current transfusion protocols are based only on platelet count thresholds. However, neonates with similar platelet counts may have different bleeding risks due to varying clinical conditions. There is an important unmet medical need to identify which neonates with low platelet counts (i.e., severe thrombocytopenia) will benefit from a transfusion. Ideally, clinicians would be able to repeatedly predict a neonate's risk of major bleeding or death with and without giving a platelet transfusion, taking into account the neonate's clinical condition at that particular time. Obtaining personalized risk estimates under specific treatment strategies, with updated predictions at each new treatment decision moment, is called 'sequential prediction under interventions'. The investigators set up an international multicenter observational cohort study to develop a model to predict major bleeding or death with and without platelet transfusion at any time point during the first week after the onset of severe thrombocytopenia. This model is designed to support platelet transfusion decisions in the NICU and may help clinicians balance the benefits and harms of platelet transfusion based on updated characteristics of the neonate at the time of prediction.

Main objective: To develop a sequential interventional prediction model to support transfusion decisions by predicting the risk of major bleeding or death with and without giving a platelet transfusion to neonates with severe thrombocytopenia (i.e., platelet count below 50x10^9/L), conditional on their characteristics present at the moment of prediction.

Study design: Multicenter international observational cohort study.

Study population: Neonates with a gestational age below 34 weeks and at least one platelet count below 50x10^9/L, admitted to a neonatal intensive care unit (NICU) between January 1st 2017 and January 1st 2022.

Main study endpoint: Major bleeding or mortality during NICU admission

Assessments: Only routine care data will be collected. This includes platelet counts and transfusions, cranial (head) ultrasounds and other information on bleeding, and multiple clinical variables.

Statistical analyses: Development of a sequential prediction model under interventions using the cloning-censoring-weighting approach with inverse probability weighting. Validation of the model in a separate cohort of preterm neonates with severe thrombocytopenia.

Inclusion Criteria:

1. Gestational age at birth <34 weeks;
2. At least one platelet count <50x109/L;
3. Admission to a participating tertiary care NICU, including postnatal transfers, between January 1st, 2017 and January 1st, 2022.

Exclusion Criteria:

1. A severe congenital malformation;
2. Only spurious platelet counts <50x109/L (e.g. clots in the sample, or a very rapid recovery to a normal platelet count without platelet transfusion);
3. Only platelet counts <50x109/L in the context of exchange transfusion;
4. Major intracranial bleeding prior to the onset of severe thrombocytopenia. Neonates with major bleeding after the end of follow-up will not be excluded, but will be recorded as having had no major bleeding during the study.

Neonates with a gestational age below 34 weeks and at least one platelet count below 50x10^9/L, admitted to a NICU between January 1st, 2017 and January 1st, 2022.

Due to the sensitivity of the data, the underlying data sets will not be shared publicly due to ethical/privacy restrictions. However, all analyses will be documented, reproducible and available for audit. The data dictionary will be available upon request or as appendix in the manuscript. We might consider restricted access to the raw data in a repository when the target journal deems this necessary for publication. For future collaboration with the research partners or for data transfer/sharing, written agreements on data management, privacy, data ownership and intellectual properties will be made, such as a research collaboration agreement and data sharing agreement, and discussed with the support of Legal/Privacy/RDM experts where appropriate. In line with the Netherlands Code of Conduct for Research Integrity, raw and processed will be stored for a period of at least 10 years. The LUMC has long-term storage with back-up available for this.