Skip to main content
NCBI home page
JAMA Network logoLink to JAMA Network
. 2025 Sep 15;334(14):1267–1277. doi: 10.1001/jama.2025.14194

Individualized Prediction of Platelet Transfusion Outcomes in Preterm Infants With Severe Thrombocytopenia

Hilde van der Staaij 1,2,3, Ilaria Prosepe 4, Camila Caram-Deelder 1, Ruth H Keogh 5, Emöke Deschmann 6, Christof Dame 7, Wes Onland 8,9, Sandra A Prins 8, Florian Cassel 10, Esther J d’Haens 11, Elke van Westering-Kroon 12, Peter Andriessen 13,14, Sabine L Vrancken 15, Christian V Hulzebos 16, Daniel C Vijlbrief 17, Suzanne F Fustolo-Gunnink 2,3,18, Karin Fijnvandraat 3,18, Enrico Lopriore 2, Johanna G van der Bom 1,✉, Nan van Geloven 4
PMCID: PMC12439188  PMID: 40952748

Key Points

Question

When and in which preterm infants with severe thrombocytopenia do platelet transfusions reduce the risk of major bleeding or death?

Findings

In a multicenter cohort of 1042 infants with severe thrombocytopenia, this study developed a decision-support algorithm estimating the absolute risks of major bleeding or mortality if prophylactic platelet transfusion were or were not to be given. Evaluation in a separate cohort (n = 637) indicated good performance.

Meaning

The findings show substantial variation in benefit and harm of platelet transfusion for severe thrombocytopenia, depending on the infant’s current condition. The algorithm’s risk estimates could help clinicians individualize prophylactic platelet transfusion decisions.

Abstract

IMPORTANCE

Preterm infants with severe thrombocytopenia (platelet count <50 × 109/L) frequently receive platelet transfusions. However, it is unclear in what cases prophylactic transfusion truly reduces bleeding risk or whether it does more harm than good.

OBJECTIVE

To develop and validate a dynamic prediction model for major bleeding or mortality if prophylactic platelet transfusion were or were not to be given to infants with severe thrombocytopenia.

DESIGN, SETTING, AND PARTICIPANTS

The dynamic prediction model was developed in an international multicenter cohort (2017-2021) comprising 14 neonatal intensive care units in the Netherlands, Sweden, and Germany. Model evaluation was performed in a national multicenter cohort (2010-2014) including 7 Dutch neonatal intensive care units. The study population consisted of infants with severe thrombocytopenia less than 34 weeks’ gestation.

EXPOSURE

Two transfusion strategies were contrasted at each prediction point: receiving a platelet transfusion within 6 hours (prophylaxis) vs no platelet transfusion for 3 days (no prophylaxis).

MAIN OUTCOMES AND MEASURES

The primary outcome was the 3-day risk of major bleeding or mortality, reestimated every 2 hours during the first week after severe thrombocytopenia onset. Predictors included gestational and postnatal age, small-for-gestational-age infant, necrotizing enterocolitis, sepsis, mechanical ventilation, vasoactive agents, platelet count, and prior platelet transfusion(s). Landmarking combined with the clone-censor-weight approach enabled dynamic prediction under the 2 transfusion strategies, accounting for time-varying confounding. Model performance was evaluated in the external validation cohort.

RESULTS

In both the development (n = 1042) and validation (n = 637) cohorts, the median gestational age was 28 weeks and median birth weight was 900 g; there were 613 (59%) and 370 (58%) males, respectively. Major bleeding or death occurred in 235 infants (23%) in the development cohort and 135 (21%) in the validation cohort. In the validation cohort, the time-dependent area under the receiver operating characteristic curve was 0.69 (95% CI, 0.60-0.76) for the prophylaxis strategy and 0.85 (95% CI, 0.76-0.92) for the no prophylaxis strategy, with calibration plots showing good calibration. Estimated risks under both strategies varied considerably depending on the infant’s clinical condition at the time of prediction.

CONCLUSIONS AND RELEVANCE

Among preterm infants with severe thrombocytopenia, this modeling study found substantial variation among individuals in predicted benefits and harms of prophylactic platelet transfusion based on their current clinical characteristics. The dynamic prediction model performed well in a validation cohort, and its value to support individualized decisions warrants evaluation in future studies.

This study examines a dynamic prediction model for major bleeding or mortality if prophylactic platelet transfusion were or were not to be given to infants with severe thrombocytopenia.

Introduction

Severe thrombocytopenia, defined as a platelet count less than 50 × 109/L (equivalent to 50 × 103/µL), occurs in 2% to 9% of infants admitted to neonatal intensive care units (NICUs) and in up to 28% of extremely preterm (<28 weeks’ gestation) infants and those with a birth weight less than 1000 g.1,2,3,4,5,6,7,8,9 These infants frequently receive platelet transfusions with the aim to prevent severe, life-threatening bleeding. This prophylactic platelet transfusion strategy is based on the assumption that increasing the platelet count will reduce the risk of bleeding, given the crucial role of platelets in primary hemostasis.

However, multiple studies have shown a weak or absent association between platelet count and bleeding in preterm infants.1,2,5,10,11,12,13 Furthermore, the PlaNeT-2/MATISSE randomized clinical trial on platelet transfusion thresholds demonstrated that using a liberal platelet count threshold of 50 × 109/L compared with a restrictive threshold of 25 × 109/L significantly increased, instead of decreased, the risk of major bleeding or death. In addition, bronchopulmonary dysplasia was more common in the liberal threshold group.14 These adverse effects of platelet transfusions persisted into childhood, with higher rates of death or neurodevelopmental impairment at 2 years of age for children previously randomized to the liberal threshold group.15 This evidence led to a strong recommendation in international guidelines to only provide prophylactic platelet transfusions for infants without signs of major bleeding when the platelet count is less than 25 × 109/L.16

Questions remain as to whether platelet transfusions ever reduce the risk of major bleeding in preterm infants and, if so, who is likely to benefit from these transfusions. Studies in adults have shown that prophylactic platelet transfusions can prevent bleeding among certain high-risk patients.17,18 In addition, individual responses to transfusion may vary, and the risk of bleeding or mortality, as well as the effect of prophylactic platelet transfusion, may change as the infant’s clinical condition evolves.

This highlights the need for an individualized approach, taking into account time-varying information on platelet counts and other factors determining an infant’s risk of bleeding, to identify infants who may benefit from transfusion vs those at risk of transfusion-related harm. Recently, methodology has been developed to train and evaluate prediction models for estimating individualized risks under different treatment strategies from observational data.19,20,21,22,23,24 The aim of this study was to develop and evaluate a decision-support algorithm to help clinicians individualize prophylactic platelet transfusion decisions in preterm infants with severe thrombocytopenia.

Methods

Data Sources

For model development, the study team conducted the Predicting Outcomes in Preterm Neonates With Thrombocytopenia (PROSPECT) study, an international longitudinal cohort study in 10 NICUs in the Netherlands, 3 in Sweden, and 1 in Germany between January 2017 and December 2021 (Supplement 1). The institutional review board of the Leiden University Medical Center in the Netherlands approved the study (protocol 22-3028) followed by institutional review boards of each center. Eight centers used an opt-out consent procedure; the other centers waived the need for informed consent. The study protocol has been previously published (NCT06043050) and the statistical analysis plan was approved by the Leiden University Medical Center Department of Clinical Epidemiology (protocol A219; Supplement 2). Model evaluation used data from the Monitoring Outcome in Neonatal Thrombocytopenia (MONET) study, a longitudinal cohort study in 7 Dutch NICUs between January 2010 and December 2014, as described elsewhere.6 The studies adhered to the Declaration of Helsinki and General Data Protection Regulation.25 The study report followed the TRIPOD+AI guidelines and included information on the causal aspects for the development and evaluation of interventional prediction models.22,26,27 Data were collected from medical records, laboratory systems, and cranial ultrasonography reports. Analyses were performed in Stata, version 16.1 (StataCorp), and R, version 4.1.17 (R Foundation).28,29,30

Population

This study included all consecutive infants who met the following criteria: born before 34 weeks’ gestation, at least 1 platelet count less than 50 × 109/L, and admission to the NICU. Exclusion criteria were severe congenital malformations, major bleeding before severe thrombocytopenia, fetal and neonatal alloimmune thrombocytopenia, and thrombocytopenia due to exchange transfusion. In the validation cohort, previous admission to another NICU was an additional exclusion criterion.

Strategies Evaluated

Two transfusion strategies were compared at each point (referred to as prediction points) at which a prophylactic platelet transfusion decision could be made: prophylaxis strategy if an infant were to receive a prophylactic platelet transfusion within 6 hours and no prophylaxis strategy if an infant were not to receive a platelet transfusion for the next 3 days. These transfusion strategies reflected real-world practice, with 75% of platelet transfusions occurring within 6 hours of pretransfusion counts. The no prophylaxis strategy was aligned with the 3-day prediction window, providing a meaningful contrast between the strategies.

Outcomes

The primary outcome was a composite of major bleeding or mortality within the 3 days following each prediction point, reflecting the short half-life of transfused platelets.31,32 Major bleeding was defined as intraventricular hemorrhage (IVH) grade 3 or any grade IVH with parenchymal periventricular hemorrhagic infarction,33 cerebellar or solitary parenchymal hemorrhage on cranial ultrasonography, subdural or epidural hemorrhage causing parenchymal compression, major pulmonary hemorrhage requiring mechanical ventilation or increased ventilatory settings, and gastrointestinal hemorrhage associated with hemodynamic instability. The secondary outcome, major bleeding or mortality within 14 days, was chosen based on the result of PlaNeT-2/MATISSE, which indicated a potential delayed adverse effect of platelet transfusion.14

Predictors

Predictors were selected a priori based on literature review and expert advice (eTable 1 in Supplement 1). Time-fixed predictors included gestational age, small-for-gestational-age infant (SGA; birth weight <10th percentile),34 and postnatal age at severe thrombocytopenia onset. Time-dependent predictors were mechanical ventilation, latest platelet count, prior platelet transfusions, necrotizing enterocolitis greater than or equal to Bell stage IIA,35 clinically suspected/proven sepsis, and vasoactive agents. Time since the first platelet count less than 50 × 109/L was included, as well as an interaction between time and SGA, because the association between SGA and bleeding may be different immediately after severe thrombocytopenia onset compared with a few days later.6 Interactions with platelet transfusion were taken into account by modeling the data separately for the prophylaxis and no prophylaxis strategies. Details about the targets of estimation (prediction estimands) are presented in eMethods 1 in Supplement 1.

Statistical Analyses

Model Development

Model development combined landmarking with the clone-censor-weight approach.24,36,37 For a detailed discussion of the method, see eMethods 2-3 in Supplement 1. In brief, the prediction points (landmarks) represented 2-hour intervals at which a new prediction was made to accurately capture changes in the infant’s condition, using all updated information about the infant up to that point. At every prediction point, infants were cloned, with one clone assigned to the prophylaxis strategy and the other to no prophylaxis. Next, infants were censored when they no longer adhered to the designated transfusion strategy (eFigure 1 in Supplement 1). For the no prophylaxis strategy, this meant that infants who received a platelet transfusion within 3 days from the prediction time point were artificially censored at the time of transfusion. For the prophylaxis strategy, infants were artificially censored if they did not receive a platelet transfusion within 6 hours. Finally, the potential bias introduced by the censoring of nonadherent infants was addressed by applying inverse probability of censoring weighting based on time-dependent confounders, as identified in a directed acyclic graph (eFigure 2 in Supplement 1).24,36 Infants were eligible for inclusion at prediction points from the first platelet count less than 50 × 109/L up to 7 days thereafter at each point their platelet count was less than 50 × 109/L. If a platelet transfusion was given, a new posttransfusion platelet count was required for inclusion in the next prediction point. If major bleeding or death occurred, infants were no longer included. Two weighted Cox proportional hazards landmark models were fitted, one for each transfusion strategy, and bootstrapping was used to obtain 95% CIs for the hazard ratios.

Model Evaluation

Prediction formulas from the developed model were used to estimate the 3-day risk of major bleeding or death under both transfusion strategies at each prediction point for each infant in the validation cohort. In this dataset, some infants who were transferred to a different NICU had censored outcomes because follow-up ended after transfer to a nonparticipating NICU. The model’s ability to assign higher risk estimates to infants who experienced major bleeding or death earlier than others was evaluated using the (cumulative/dynamic) time-dependent area under the receiver operating characteristic curve (AUC) and C index. Time-dependent AUC expresses the proportion of correctly ordered risks between infants who did and who did not (yet) experience major bleeding or death by the 3-day prediction window, whereas the C index expresses the proportion of correctly ordered risks for all patient pairs with at least 1 (and possibly both) having their event before 3 days.38 To evaluate how well the estimated absolute risks corresponded to observed outcomes, calibration was assessed using the observed-to-expected ratio and calibration plots. Overall prediction accuracy was assessed using the (scaled) Brier score, which combines both calibration and discrimination.39 Specialized techniques were used to estimate these performance metrics under both (counterfactual) treatment strategies, using newly estimated time-dependent inverse probability of censoring weights derived from the validation dataset.24 Bootstrapping was used to construct 95% CIs for the performance metrics.

Results

Development and Validation Cohort Characteristics

In the development cohort, 1226 of 13101 infants (9.4%) born at less than 34 weeks’ gestation had confirmed severe thrombocytopenia. After applying the exclusion criteria, 1042 infants remained in the development cohort. In the validation cohort, severe thrombocytopenia was observed in 830 of 9333 infants (8.9%), of whom 637 were included (Figure 1). In both cohorts, the median gestational age was 28 weeks, birth weight was 900 g, and severe thrombocytopenia onset was 4 days after birth. Other demographics were also similar, except for a higher rate of multiple platelet transfusions in the validation cohort (Table).

Figure 1. Flow of the Development and Validation Cohorts.

Figure 1.

Open in a new tab

Development cohort: PROSPECT study (2017-2021). Validation cohort: MONET study (2010-2014). NICU indicates neonatal intensive care unit.

aSpurious platelet count due to clots, rapid spontaneous recovery, or labeled as laboratory error.

bThe composite outcome of major bleeding or death within 21 days of the first platelet count <50 × 109/L (final prediction point on day 7 plus the 14-day prediction window for the secondary outcome) was estimated using the Kaplan-Meier method to account for censoring due to transfer of infants to a nonparticipating NICU in the validation cohort (see eResults 1 in Supplement 1).

cThe number of infants receiving at least one platelet transfusion during the entire follow-up up period of 21 days. The median (IQR) number of platelet transfusions received in transfused infants was 2 (1-3) in both cohorts.

dInfants may contribute to multiple prediction points (2-hour landmarks from the onset of severe thrombocytopenia up to 7 days), using the infant’s updated information at each new prediction point. Because of the cloning step of the clone-censor-weight approach, both transfusion strategies have the same number of predictions (see eMethods 3 in Supplement 1).

Table. Cohort Characteristics, Covariates, and Outcomes.

Cohort characteristics Cohort
Development Validation
At birth
No. of infants 1042 637
Gestational age, median (IQR), wka 28 (26-30) 28 (26-30)
Male sex, No. (%) 613 (59) 370 (58)
Female sex, No. (%) 429 (41) 267 (42)
Birth weight, median (IQR), g 900 (698-1230) 900 (710-1177)
Small for gestational age (birth weight <10th percentile), No. (%) 601 (58) 361 (57)
Cesarean section, No. (%) 762 (73) 450 (71)
Multifetal pregnancy, No. (%) 217 (21) 162 (25)
Complete course of antenatal corticosteroids, No. (%)b 591 (57) 409 (64)
Apgar score <5 at 5 min, No. (%)c 83 (8) 64 (10)
Perinatal asphyxia, No. (%)d 76 (7) 53 (8)
Born outside the NICU center of admission, No. (%) 115 (11) 69 (11)
At the onset of severe thrombocytopenia
No. of infants 1042 637
Postnatal age, median (IQR), d 4 (2-8) 4 (2-9)
Platelet count, median (IQR), ×109/L 38 (28-45) 38 (29-45)
1 or more platelet transfusions before the first platelet count <50 × 109/L, No. (%) 51 (5) 68 (11)
At the start of each prediction point
No. of predictionse 19 910 13 846
Predictions per infant, median (IQR) 12 (4-28) 14 (5-34)
Postnatal age, median (IQR), d 6 (4-12) 9 (4-15)
Platelet count, median (IQR), ×109/L 39 (30-45) 36 (27-44)
No. (%) of prior platelet transfusion(s)
0 13 141 (66) 6930 (50)
1 3378 (17) 2525 (18)
2 1572 (8) 1806 (13)
≥3 1819 (9) 2585 (19)
Sepsis, No. (%) 10 798 (54) 8528 (62)
Mechanical ventilation, No. (%) 8130 (41) 6619 (48)
Necrotizing enterocolitis, No. (%) 3358 (17) 1913 (14)
Receiving vasoactive agents, No. (%) 2518 (13) 1713 (12)
Planned surgery or lumbar puncture, No. (%)f 446 (2.2) 331 (2.4)
Intravenous bolus, No. (%)g 80 (0.4) NA
Pharmacologic treatment for hsPDA, No. (%) 36 (0.2) 182 (1.3)
Outcomesh
No. of infants 1042 637
Major bleeding or death (composite outcome), No. (%) 258 (24.8) 156 (25.0)
Death, No. (%) 143 (13.6) 90 (14.1)
First episode of major bleeding, No. (%) 116 (11.9) 66 (10.4)
Open in a new tab

Abbreviations: hsPDA, hemodynamically significant patent ductus arteriosus treated with ibuprofen or indomethacin; NA, not available.

a

Gestational age ranged from 22 to 34 weeks’ gestation in the development cohort and from 24 to 34 weeks’ gestation in the validation cohort.

b

At least 2 doses before delivery. These data were missing for 48 infants (4.6%) in the development cohort and for 21 (3.3%) in the validation cohort.

c

The Apgar score is a standardized assessment of a newborn’s status 1, 5, and 10 minutes after birth based on 5 characteristics (color, heart rate, reflexes, muscle tone, and respiration), each assigned a value of 0 to 2. An Apgar score <5 at 5 minutes is associated with a higher risk of neonatal mortality.

d

Perinatal asphyxia defined by at least 3 criteria: 5-minute Apgar score ≤5, arterial pH <7.0, base excess <−16 mmol/L, lactate >10 mmol/L within 1 hour of birth, ≥10 minutes of respiratory resuscitation, and/or cardiopulmonary resuscitation. To convert lactate to mg/dL, divide values by 0.111.

e

Infants may contribute to multiple prediction points (ie, 2-hour landmarks from the onset of severe thrombocytopenia up to 7 days), using the infant’s updated information at each new prediction point.

f

Planned surgery in the next 6 hours and planned lumbar puncture in the next 2 hours of the prediction point.

g

Intravenous bolus received within the last hour of the prediction point for the management of (suspected) hypotension. This variable was not collected in the validation cohort.

h

The composite outcome of major bleeding or death within 21 days of the first platelet count <50 × 109/L (final prediction point on day 7 plus the 14-day prediction window for the secondary outcome), as well as the individual components (death and major bleeding separately), were estimated using the Kaplan-Meier method to account for censoring due to transfer of infants to a nonparticipating NICU in the validation cohort (see eResults 1 in Supplement 1).

There was minimal variation in the probability of receiving a platelet transfusion between centers in the development cohort. In the validation cohort, 1 center showed a distinct transfusion pattern, which was adjusted by weighting (eFigure 3 in Supplement 1). Platelet transfusions in the development cohort were administered with a median (IQR) volume of 14 (10-15) mL/kg , duration of 30 (30-60) minutes, and infusion rate of 20 (15-30) mL/kg per hour. In the validation cohort, the median (IQR) volume was 10 (10-15) mL/kg, with no information on duration available (eMethods 3 in Supplement 1). Because completeness of data was carefully monitored during data collection, there were no missing data for measured predictors or confounders.

Major Bleeding and Mortality

Major bleeding or death occurred in 23% of infants (235 of 1042) within 10 days of the first platelet count less than 50 × 109/L in the development cohort, including 108 episodes of major bleeding and 127 deaths. In the validation cohort, 21% of infants (135 of 637) experienced major bleeding or death, with 63 episodes of major bleeding and 72 deaths. Grade 3 IVH was the most common type of major bleeding in both cohorts, followed by major pulmonary and cerebellar hemorrhage (eResults 1 in Supplement 1). Nearly all patients underwent cranial ultrasonography at least once after severe thrombocytopenia onset: 98% (1024 of 1042) in the development cohort and 95% (604 of 637) in the validation cohort. eResults 2 in Supplement 1 describe the number of infants and predictions corresponding to an event, discharge, or transfer by transfusion strategy.

Dynamic Model and Model Performance

Figure 2 reports the hazard ratios of the predictors in the development cohort. All validation measures of the model’s predictive performance were assessed in the validation cohort. Figure 3 shows the calibration plot of the model. The median (IQR) predicted 3-day risk of major bleeding or mortality was 7.4% (4.2%-14.3%) under the prophylaxis strategy and 6.0% (2.6%-16.3%) under the no prophylaxis strategy (eResults 3 in Supplement 1). For both strategies, estimated risks were similar to the observed outcomes, with observed-to-expected ratios of 1.01 (95% CI, 0.73-1.32) for the prophylaxis strategy and 0.96 (95% CI, 0.51-1.58) for the no prophylaxis strategy, although risks were somewhat overestimated for the highest 20% of predicted values under both strategies. The time-dependent AUC was 0.69 (95% CI, 0.60-0.76) for the prophylaxis strategy and 0.85 (95% CI, 0.76-0.92) for the no prophylaxis strategy. Corresponding C indexes were 0.75 (95% CI, 0.69-0.80) and 0.83 (95% CI, 0.72-0.89), respectively.

Figure 2. Strength of the Predictors for Major Bleeding or Mortality in the Development Cohort.

Figure 2.

Open in a new tab

A hazard ratio (HR) >1.0 indicates that an increase in the value of the predictor is associated with a higher risk of major bleeding or death within 14 days, and an HR <1.0 indicates a lower risk of this outcome. 95% CIs were obtained via bootstrapping using 500 bootstrapped samples. It is important to note that the HRs of the predictors in the model should not be interpreted as representing their causal effect on the outcome. The aim of the study is not to estimate the causal effect of each individual predictor, but rather to accurately predict the risk of major bleeding or mortality under the 2 platelet transfusion strategies, using the combination of all predictors in the model. Details of the full prediction formulas are shown in eResults 6 in Supplement 1. SGA indicates small for gestational age.

aProphylaxis strategy: if an infant were to receive a prophylactic platelet transfusion within 6 hours of the prediction point.

bNo prophylaxis strategy: if an infant were not to receive a prophylactic platelet transfusion within the next 3 days of the prediction point.

cLandmark time is the prediction point, ie, the time in hours since the first platelet count <50 × 109/L.

Figure 3. Distribution of Risk Estimates by Transfusion Strategy.

Figure 3.

Open in a new tab

AUCt indicates the time-dependent area under the receiver operating characteristic curve, indicating the proportion of correctly ordered risks among infants who did or did not (yet) experience major bleeding or death within 3 days. C index reflects the proportion of correctly ordered risks among all infant pairs in which at least 1 (and possibly both) had an event before 3 days. Observed-to-expected (OE) ratio indicates how well the estimated absolute risks corresponded to observed outcomes; values <1 suggest overestimation and >1 suggest underestimation. The dots represent quintiles of the predictions according to the predicted values, with the mean prediction of each quintile on the x-axis and the mean observed outcome proportion of that group on the y-axis, including 95% CIs obtained by bootstrapping using 500 bootstrap samples. Error bars indicate 95% CIs. The histograms along the x-axis show the distribution of risk estimates for both transfusion strategies.

The Brier score was 0.02 (95% CI, 0.02-0.03) for the prophylaxis strategy and 0.07 (95% CI, 0.04-0.09) for the no prophylaxis strategy. Scaled Brier scores were 15.3% (95% CI, 9.2%-21.9%) for the prophylaxis strategy and 20.9% (95% CI, −16.3% to 52.4%) for the no prophylaxis strategy, suggesting better performance than a null model without covariates fitted to the validation data, which estimates under each strategy the mean risk for all infants. The wider CI and negative lower bound for the no prophylaxis strategy indicated more statistical uncertainty. The performance metrics for the 14-day prediction window and mean estimated/observed survival curves are reported in eResults 3-5 in Supplement 1.

Predictions of the Dynamic Model

eFigure 7 in Supplement 1 illustrates the use of the model to repeatedly predict major bleeding or mortality in the next 3 days if an infant were to receive prophylactic platelet transfusion (prophylaxis strategy) and the risk if they were not to receive prophylactic platelet transfusion (no prophylaxis strategy). Figure 4A shows the model’s risk estimates for example patients based on data from the development dataset. It demonstrates that an infant’s risk of major bleeding or death can vary between the strategies and that infants with the same platelet count may have different risks depending on other clinical characteristics. The differences in estimated risks between the strategies help illustrate individualized platelet transfusion effects. The full characteristics of the example patients are described in eTable 4 in Supplement 1.

Figure 4. Illustrations of Predictions From the Dynamic Interventional Prediction Model.

Figure 4.

Open in a new tab

A, For example patients 1 and 2, a platelet count of 25 × 109/L was selected to reflect the current threshold for prophylactic platelet transfusion used in the participating centers. All example patients have a gestational age of 28 weeks. For example patients 3 and 4, a higher platelet count of 35 × 109/L was chosen, but otherwise they have similar characteristics as patients 1 and 2, respectively. For better visualization, the upper limit of the 95% CI of example patient 1 (ranging from 35%-49%) was truncated at 35% from day 2. The full characteristics of the example patients are described in eTable 4 and the supplemental prediction tool in Supplement 1. B and C, Grey dots show the risk differences. Boxes indicate 25th to 75th percentiles, horizontal lines indicate the median, and whiskers the 5th to 95th percentiles. The orange line represents no risk difference; values >0 indicate potential benefit of transfusion and values <0 indicate potential harm. The estimated values for example patients 1-4 are shown as orange dots. NEC indicates necrotizing enterocolitis.

Figure 4B shows the risk under no prophylaxis minus the risk under prophylaxis against platelet count at each prediction point in the development dataset. These estimated risk differences vary across the whole platelet count range, indicating benefit of transfusion (lower risk under prophylaxis strategy) for most predictions with platelet counts less than 20 × 109/L, and more frequently a harmful effect at higher platelet counts. Figure 4C shows the estimated risk difference against the estimated risk under the no prophylaxis strategy. Overall, when the estimated untreated risk is low (<5%), most predictions suggest harm from transfusion, whereas as the untreated risk increases, more predictions suggest benefit from transfusion. The risk differences between strategies plotted against each predictor in the model are shown in eFigure 8 in Supplement 1.

Discussion

This study developed and validated a dynamic prediction model to estimate individualized absolute risks of major bleeding or death both with and without a subsequent prophylactic platelet transfusion in preterm infants less than 34 weeks’ gestation with severe thrombocytopenia. The findings show substantial variation of potential beneficial and harmful effects of platelet transfusion depending on the infant’s current clinical condition. This is, to the best of the authors’ knowledge, the first study to develop a decision-support algorithm for prophylactic platelet transfusion decisions in preterm infants.

Strengths of this study include a large development cohort of more than 1000 infants with severe thrombocytopenia. Confounders and readily obtainable predictors were selected before analyses and data collection was meticulous. Evaluation of the model in a cohort distinct from the development dataset showed good performance over the 3-day prediction window. The calibration plot showed good overall agreement between predicted and observed outcomes for both the prophylaxis and no prophylaxis strategies, with observed-to-expected ratios close to 1, although the 95% CIs were wide. For the highest 20% of predicted values, the model somewhat overestimated the observed risks for both strategies. This overestimation was similar for both strategies, suggesting that the difference between the estimated risks may be correctly estimated.

The current study’s findings align with those of PlaNeT-2/MATISSE, which demonstrated that prophylactic platelet transfusions given above a platelet count threshold of 25 × 109/L were associated with a higher risk of major bleeding or death.14 Secondary analysis of the trial showed benefit from the restrictive threshold, regardless of estimated risks of bleeding or death.40 However, the secondary analysis modeled risks using data available at trial inclusion. It did not use updated (dynamic) information on changes in an infant’s clinical condition, and it was not designed to estimate risks if no platelet transfusion were to be administered.

This study had 3 novel aspects. First, the model enables prediction of major bleeding or death at any point during the first week after severe thrombocytopenia onset for 2 potential (counterfactual) scenarios: if an infant were to receive a prophylactic platelet transfusion and if an infant were not to receive transfusion. The difference between these risks indicates the estimated individualized causal effect of prophylactic platelet transfusion.21 Second, these predicted effects showed considerable variation across different platelet counts, suggesting that there are other factors besides platelet count that influence the effect of platelet transfusion. Thus, a more elaborate, individualized transfusion approach incorporating multiple clinical variables, such as based on this decision-support algorithm, may provide better guidance to prophylactic platelet transfusion decisions than a single platelet count threshold. Third, these findings show that prophylactic platelet transfusions were associated with less major bleeding or death in some thrombocytopenic infants. This is biologically plausible but has not previously been supported by quantitative evidence.41,42

Future studies should investigate whether incorporating the model’s risk estimates into clinical decision-making improves outcomes, ideally through randomized impact trials comparing model-guided transfusion decisions with usual care.43,44 In preparation for such trials, expert consensus could define clinically relevant cutpoints in which the estimated risks of bleeding or mortality are meaningfully lower under prophylaxis than under no prophylaxis. Trial emulations using observational data can assess the potential impact of these thresholds as well as identify subgroups (eg, by gestational age, postnatal age, or SGA) where the model has the greatest potential to reduce unnecessary platelet transfusions. In different geographical settings, additional validation studies may be needed.45,46 Before evidence on the model’s impact becomes available, it is recommended to use the restrictive platelet count threshold of 25 × 109/L in line with international guidelines.14,16,40

Limitations

This study has several limitations. First, platelet transfusion practice was more liberal in the validation cohort (2010-2014) than in the development cohort (2017-2021) due to guideline changes following the PlaNeT-2/MATISSE publication, which may have affected calibration.45,46 Second, it was not possible to predict major bleeding or mortality separately due to sample size limitations. Third, residual confounding due to unknown or unmeasured factors cannot be ruled out. Fourth, because the majority of the study population was White, it was not possible to evaluate model performance across diverse racial and ethnic groups. Fifth, similar to any prediction model, predictions based on a limited sample carry uncertainty.47,48

Sixth, confirming that all platelet counts were measured before major bleeding is challenging, because IVH is often asymptomatic and detected with time delay by routine cranial ultrasonography screening. Seventh, platelet transfusion was modeled as a binary intervention reflecting real-world practice, with a median volume of circa 15 mL/kg over 30 minutes and an infusion rate of 20 mL/kg per hour, representative for European NICUs.49 However, it has been suggested that hyperconcentrated platelet products (circa 3 mL/kg) induce less fluctuation in intravascular volume and blood pressure due to rapid volume expansion, potentially decreasing IVH risk.50,51,52 This model would need to be adjusted if future studies find that transfusing smaller volumes or at slower rates reduces bleeding risk.41,53,54

Eighth, in the future, other parameters may improve bleeding prediction, such as measures of platelet production, reflected by immature platelet fraction, and measures of primary hemostasis, such as closure time and viscoelastic coagulation tests.12,55,56,57,58,59,60 These parameters were currently not included because they are not yet commonly used in clinical practice and their value in clinical decision-making should be further established.

Conclusions

Among severely thrombocytopenic preterm infants, this modeling study found substantial variation among individuals in predicted benefits and harms of prophylactic platelet transfusion at any given platelet count less than 50 × 109/L based on their current clinical characteristics. The dynamic prediction model had good performance in a validation cohort, and its value to support individualized platelet transfusion decisions warrants evaluation in future studies.

Supplement 1.

eMethods

jama-e2514194-s001.pdf (2.6MB, pdf)
Supplement 2.

Statistical analysis plan

jama-e2514194-s002.pdf (3.4MB, pdf)
Supplement 3.

eTable

jama-e2514194-s003.xlsx (70.9KB, xlsx)
Supplement 4.

Data sharing statement

jama-e2514194-s004.pdf (13.5KB, pdf)

References

  • 1.Baer VL, Lambert DK, Henry E, Christensen RD. Severe thrombocytopenia in the NICU. Pediatrics. 2009;124(6):e1095-e1100. doi: 10.1542/peds.2009-0582 [DOI] [PubMed] [Google Scholar]
  • 2.Stanworth SJ, Clarke P, Watts T, et al. ; Platelets and Neonatal Transfusion Study Group . Prospective, observational study of outcomes in neonates with severe thrombocytopenia. Pediatrics. 2009;124(5):e826-e834. doi: 10.1542/peds.2009-0332 [DOI] [PubMed] [Google Scholar]
  • 3.von Lindern JS, van den Bruele T, Lopriore E, Walther FJ. Thrombocytopenia in neonates and the risk of intraventricular hemorrhage: a retrospective cohort study. BMC Pediatr. 2011;11:16. doi: 10.1186/1471-2431-11-16 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Murray NA, Howarth LJ, McCloy MP, Letsky EA, Roberts IA. Platelet transfusion in the management of severe thrombocytopenia in neonatal intensive care unit patients. Transfus Med. 2002;12(1):35-41. doi: 10.1046/j.1365-3148.2002.00343.x [DOI] [PubMed] [Google Scholar]
  • 5.van der Staaij H, Hooiveld NMA, Caram-Deelder C, et al. Most major bleeds in preterm infants occur in the absence of severe thrombocytopenia: an observational cohort study. Arch Dis Child Fetal Neonatal Ed. 2024:110(2):122-127. doi: 10.1136/archdischild-2024-326959 [DOI] [PubMed] [Google Scholar]
  • 6.Fustolo-Gunnink SF, Fijnvandraat K, Putter H, et al. Dynamic prediction of bleeding risk in thrombocytopenic preterm neonates. Haematologica. 2019;104(11):2300-2306. doi: 10.3324/haematol.2018.208595 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Christensen R, Henry E, Wiedmeier S, et al. Thrombocytopenia among extremely low birth weight neonates: data from a multihospital healthcare system. J Perinatol. 2006;26(6):348-353. doi: 10.1038/sj.jp.7211509 [DOI] [PubMed] [Google Scholar]
  • 8.Bolat F, Kılıç SÇ, Oflaz MB, et al. The prevalence and outcomes of thrombocytopenia in a neonatal intensive care unit: a three-year report. Pediatr Hematol Oncol. 2012;29(8):710-720. doi: 10.3109/08880018.2012.725454 [DOI] [PubMed] [Google Scholar]
  • 9.von Lindern JS, Hulzebos CV, Bos AF, Brand A, Walther FJ, Lopriore E. Thrombocytopaenia and intraventricular haemorrhage in very premature infants: a tale of two cities. Arch Dis Child Fetal Neonatal Ed. 2012;97(5):F348-F352. doi: 10.1136/fetalneonatal-2011-300763 [DOI] [PubMed] [Google Scholar]
  • 10.Sparger KA, Assmann SF, Granger S, et al. Platelet transfusion practices among very-low-birth-weight infants. JAMA Pediatr. 2016;170(7):687-694. doi: 10.1001/jamapediatrics.2016.0507 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Muthukumar P, Venkatesh V, Curley A, et al. ; Platelets Neonatal Transfusion Study Group . Severe thrombocytopenia and patterns of bleeding in neonates: results from a prospective observational study and implications for use of platelet transfusions. Transfus Med. 2012;22(5):338-343. doi: 10.1111/j.1365-3148.2012.01171.x [DOI] [PubMed] [Google Scholar]
  • 12.Deschmann E, Saxonhouse MA, Feldman HA, Norman M, Barbian M, Sola-Visner M. Association of bleeding scores and platelet transfusions with platelet counts and closure times in response to adenosine diphosphate (CT-ADPs) among preterm neonates with thrombocytopenia. JAMA Netw Open. 2020;3(4):e203394. doi: 10.1001/jamanetworkopen.2020.3394 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Resch E, Hinkas O, Urlesberger B, Resch B. Neonatal thrombocytopenia-causes and outcomes following platelet transfusions. Eur J Pediatr. 2018;177(7):1045-1052. doi: 10.1007/s00431-018-3153-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Curley A, Stanworth SJ, Willoughby K, et al. ; PlaNeT2 MATISSE Collaborators . Randomized Trial of platelet-transfusion thresholds in neonates. N Engl J Med. 2019;380(3):242-251. doi: 10.1056/NEJMoa1807320 [DOI] [PubMed] [Google Scholar]
  • 15.Moore CM, D’Amore A, Fustolo-Gunnink S, et al. ; PlaNeT2 MATISSE . Two-year outcomes following a randomised platelet transfusion trial in preterm infants. Arch Dis Child Fetal Neonatal Ed. 2023;108(5):452-457. doi: 10.1136/archdischild-2022-324915 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Metcalf RA, Nahirniak S, Guyatt G, et al. Platelet transfusion: 2025 AABB and ICTMG international clinical practice guidelines. JAMA. 2025;334(7):606-617. doi: 10.1001/jama.2025.7529 [DOI] [PubMed] [Google Scholar]
  • 17.Stanworth SJ, Estcourt LJiyusa, Powter G, et al. ; TOPPS Investigators . A no-prophylaxis platelet-transfusion strategy for hematologic cancers. N Engl J Med. 2013;368(19):1771-1780. doi: 10.1056/NEJMoa1212772 [DOI] [PubMed] [Google Scholar]
  • 18.Stanworth SJ, Hudson CL, Estcourt LJ, Johnson RJ, Wood EM. Risk of bleeding and use of platelet transfusions in patients with hematologic malignancies: recurrent event analysis. Haematologica. 2015;100(6):740-747. doi: 10.3324/haematol.2014.118075 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Dickerman BA, Hernán MA. Counterfactual prediction is not only for causal inference. Eur J Epidemiol. 2020;35(7):615-617. doi: 10.1007/s10654-020-00659-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.van Geloven N, Swanson SA, Ramspek CL, et al. Prediction meets causal inference: the role of treatment in clinical prediction models. Eur J Epidemiol. 2020;35(7):619-630. doi: 10.1007/s10654-020-00636-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Nguyen TL, Collins GS, Landais P, Le Manach Y. Counterfactual clinical prediction models could help to infer individualized treatment effects in randomized controlled trials: an illustration with the International Stroke Trial. J Clin Epidemiol. 2020;125:47-56. doi: 10.1016/j.jclinepi.2020.05.022 [DOI] [PubMed] [Google Scholar]
  • 22.Luijken K, Morzywołek P, van Amsterdam W, et al. Risk-based decision making: estimands for sequential prediction under interventions. Biom J. 2024;66(8):e70011. doi: 10.1002/bimj.70011 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Boyer CCB, Dahabreh IJ, Steingrimsson JA. Estimating and evaluating counterfactual prediction models. arXiv. Preprint posted August 12, 2025. doi: 10.48550/arXiv.2308.13026 [DOI]
  • 24.Keogh RH, Van Geloven N. Prediction under interventions: evaluation of counterfactual performance using longitudinal observational data. Epidemiology. 2024;35(3):329-339. doi: 10.1097/EDE.0000000000001713 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.World Medical Association . World Medical Association Declaration of Helsinki: ethical principles for medical research involving human participants. JAMA. 2025;333(1):71-74. doi: 10.1001/jama.2024.21972 [DOI] [PubMed] [Google Scholar]
  • 26.Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi: 10.1136/bmj-2023-078378 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.van Amsterdam W, Cinà G, Didelez V, et al. Rapid response to: TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi: 10.1136/bmj-2023-078378 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Castor Electronic Data Capture System . Accessed October 24, 2024. https://www.castoredc.com/
  • 29.R Development Core Team . R: A Language and Environment for Statistical Computing, version 4.3.1. R Foundation for Statistical Computing; 2023. [Google Scholar]
  • 30.StataCorp . Stata Statistical Software: Release 16. StataCorp LLC; 2019. [Google Scholar]
  • 31.Moore CM, Curley A. Platelet transfusion thresholds in neonatal medicine. Early Hum Dev. 2019;138:104845. doi: 10.1016/j.earlhumdev.2019.104845 [DOI] [PubMed] [Google Scholar]
  • 32.Lambert MP, Sullivan SK, Fuentes R, French DL, Poncz M. Challenges and promises for the development of donor-independent platelet transfusions. Blood. 2013;121(17):3319-3324. doi: 10.1182/blood-2012-09-455428 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Volpe JJ. Intraventricular hemorrhage in the premature infant–current concepts: part II. Ann Neurol. 1989;25(2):109-116. doi: 10.1002/ana.410250202 [DOI] [PubMed] [Google Scholar]
  • 34.Hoftiezer L, Hof MHP, Dijs-Elsinga J, Hogeveen M, Hukkelhoven CWPM, van Lingen RA. From population reference to national standard: new and improved birthweight charts. Am J Obstet Gynecol. 2019;220(4):383.e1-383.e17. doi: 10.1016/j.ajog.2018.12.023 [DOI] [PubMed] [Google Scholar]
  • 35.Bell MJ, Ternberg JL, Feigin RD, et al. Neonatal necrotizing enterocolitis: therapeutic decisions based upon clinical staging. Ann Surg. 1978;187(1):1-7. doi: 10.1097/00000658-197801000-00001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Hernán MA. How to estimate the effect of treatment duration on survival outcomes using observational data. BMJ. 2018;360:k182. doi: 10.1136/bmj.k182 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Van Houwelingen HC. Dynamic prediction by landmarking in event history analysis. Scand J Stat Theory Appl. 2007;34(1):70-85. doi: 10.1111/j.1467-9469.2006.00529.x [DOI] [Google Scholar]
  • 38.Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics. 2005;61(1):92-105. doi: 10.1111/j.0006-341X.2005.030814.x [DOI] [PubMed] [Google Scholar]
  • 39.Graf E, Schmoor C, Sauerbrei W, Schumacher M. Assessment and comparison of prognostic classification schemes for survival data. Stat Med. 1999;18(17-18):2529-2545. doi: [DOI] [PubMed] [Google Scholar]
  • 40.Fustolo-Gunnink SF, Fijnvandraat K, van Klaveren D, et al. ; PlaNeT2 and MATISSE collaborators . Preterm neonates benefit from low prophylactic platelet transfusion threshold despite varying risk of bleeding or death. Blood. 2019;134(26):2354-2360. doi: 10.1182/blood.2019000899 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Moore CM, Curley AE. Neonatal platelet transfusions: starting again. Transfus Med Rev. 2021;35(3):29-35. doi: 10.1016/j.tmrv.2021.06.003 [DOI] [PubMed] [Google Scholar]
  • 42.Fustolo-Gunnink SF, Huisman EJ, van der Bom JG, et al. Are thrombocytopenia and platelet transfusions associated with major bleeding in preterm neonates? a systematic review. Blood Rev. 2019;36:1-9. doi: 10.1016/j.blre.2018.10.001 [DOI] [PubMed] [Google Scholar]
  • 43.Sachs MC, Sjölander A, Gabriel EE. Aim for clinical utility, not just predictive accuracy. Epidemiology. 2020;31(3):359-364. doi: 10.1097/EDE.0000000000001173 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.de Hond AAH, Vosslamber S, Lange S, et al. When, where, who, what, and why? the five Ws of workflow analysis for implementing an AI decision support tool at the intensive care. Hum Factors Healthc. 2025;7:100095. doi: 10.1016/j.hfh.2025.100095 [DOI] [Google Scholar]
  • 45.Van Calster B, Steyerberg EW, Wynants L, van Smeden M. There is no such thing as a validated prediction model. BMC Med. 2023;21(1):70. doi: 10.1186/s12916-023-02779-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW; Topic Group ‘Evaluating diagnostic tests and prediction models’ of the STRATOS initiative . Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230. doi: 10.1186/s12916-019-1466-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Thomassen D, le Cessie S, van Houwelingen HC, Steyerberg EW. Effective sample size: a measure of individual uncertainty in predictions. Stat Med. 2024;43(7):1384-1396. doi: 10.1002/sim.10018 [DOI] [PubMed] [Google Scholar]
  • 48.Riley RD, Collins GS, Kirton L, et al. Uncertainty of risk estimates from clinical prediction models: rationale, challenges, and approaches. BMJ. 2025;388:e080749. doi: 10.1136/bmj-2024-080749 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Houben NAM, Lopriore E, Fijnvandraat K, et al. Platelet transfusion in neonatal intensive care units of 22 European countries: a prospective observational study. Lancet Reg Health Eur. 2024;47:101086. doi: 10.1016/j.lanepe.2024.101086 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Scrivens A, Reibel NJ, Heeger L, et al. Survey of transfusion practices in preterm infants in Europe. Arch Dis Child Fetal Neonatal Ed. 2023:108(4):360-366. doi: 10.1136/archdischild-2022-324619 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Ferrer-Marín F, Sola-Visner M. Neonatal platelet physiology and implications for transfusion. Platelets. 2022;33(1):14-22. doi: 10.1080/09537104.2021.1962837 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Honohan A, van’t Ende E, Hulzebos C, et al. Posttransfusion platelet increments after different platelet products in neonates: a retrospective cohort study. Transfusion. 2013;53(12):3100-3109. doi: 10.1111/trf.12127 [DOI] [PubMed] [Google Scholar]
  • 53.Dannaway DC, Noori S. A randomized trial of platelet transfusions over 30 vs 120 minutes: is there an effect on post-transfusion platelet counts? J Perinatol. 2013;33(9):703-706. doi: 10.1038/jp.2013.46 [DOI] [PubMed] [Google Scholar]
  • 54.Davenport P, Sola-Visner M. Recent advances in NICU platelet transfusions. Semin Fetal Neonatal Med. 2025;30(1):101609. doi: 10.1016/j.siny.2025.101609 [DOI] [PubMed] [Google Scholar]
  • 55.The Neonatal Hemorrhagic Risk Assessment in Thrombocytopenia (NEOHAT-2). ClinicalTrials.gov. Accessed May 30, 2023. https://clinicaltrials.gov/study/NCT04598750
  • 56.Deschmann E, Saxonhouse MA, Feldman HA, Norman M, Barbian M, Sola-Visner M. Association between in vitro bleeding time and bleeding in preterm infants with thrombocytopenia. JAMA Pediatr. 2019;173(4):393-394. doi: 10.1001/jamapediatrics.2019.0008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Cremer M, Paetzold J, Schmalisch G, et al. Immature platelet fraction as novel laboratory parameter predicting the course of neonatal thrombocytopenia. Br J Haematol. 2009;144(4):619-621. doi: 10.1111/j.1365-2141.2008.07485.x [DOI] [PubMed] [Google Scholar]
  • 58.MacQueen BC, Christensen RD, Henry E, et al. The immature platelet fraction: creating neonatal reference intervals and using these to categorize neonatal thrombocytopenias. J Perinatol. 2017;37(7):834-838. doi: 10.1038/jp.2017.48 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Raffaeli G, Tripodi A, Cavallaro G, et al. Thromboelastographic profiles of healthy very low birthweight infants serially during their first month. Arch Dis Child Fetal Neonatal Ed. 2020;105(4):412-418. doi: 10.1136/archdischild-2019-317860 [DOI] [PubMed] [Google Scholar]
  • 60.Radicioni M, Bruni A, Bini V, Villa A, Ferri C. Thromboelastographic profiles of the premature infants with and without intracranial hemorrhage at birth: a pilot study. J Matern Fetal Neonatal Med. 2015;28(15):1779-1783. doi: 10.3109/14767058.2014.968773 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

eMethods

jama-e2514194-s001.pdf (2.6MB, pdf)
Supplement 2.

Statistical analysis plan

jama-e2514194-s002.pdf (3.4MB, pdf)
Supplement 3.

eTable

jama-e2514194-s003.xlsx (70.9KB, xlsx)
Supplement 4.

Data sharing statement

jama-e2514194-s004.pdf (13.5KB, pdf)

Articles from JAMA are provided here courtesy of American Medical Association

RESOURCES