The study design and data acquisition were approved by the institutional review board of the University Heidelberg, Germany, No S-721/2018, and the study was performed in accordance with the Declaration of Helsinki.

Patients undergoing ASCT for multiple myeloma at the in-patient service of the University Hospital Heidelberg between 2011 and February 2020 were included. Patients undergoing their ASCT as outpatients, being on dual APT or therapeutic anticoagulation, were excluded. Patient characteristics and the course of ASCT were captured by the review of electronical medical records (EMRs).

The primary outcome was the occurrence of BLEDs during hospital stay. These were staged according to CTCAE, version 5.0 (National Cancer Institute, Bethesda, MD), the Bleeding Severity Measurement Scale (BSMS) (25), and the World Health Organization (WHO) bleeding scale. In-depth discussion and rationale for the use of these bleeding scales are summarized and discussed in the Supplementary Material (1.1 Supplementary Methods , 1.1.1 Grading of bleeding events, their definitions, and rationale).

Secondary outcomes included the number of PLTXs, the time to the engraftment of platelets, and MACE. For the purpose of this study, MACE was defined as nonfatal stroke, nonfatal myocardial infarction, and CV death.

ASA therapy at a dose of 75–100 mg during ASCT was managed at individual physicians’ discretion. ASA was regarded as having an efficient antiplatelet function during thrombocytopenia if at least one dose of ASA after the reinfusion of stem cells (day 0) was provided as the autologous stem-cell preparation contained platelets and plasma potentially reverting the ASA effect. Furthermore, it was assumed that all platelets at this point were sufficiently inhibited and no turnover took place until hematopoietic recovery. ASA was usually recommenced with the recovery of PLC >20–50 × 10⁹/L. Patients who stopped ASA intake between hospital admission and earlier than 1 day after stem-cell reinfusion (day +1) were excluded from main analyses because neither ASA efficacy could not be assumed nor excluded. BLEDs and the patient characteristics of this group were separately assessed and analyzed to exclude a selection bias.

During the study period, two major transfusion strategies existed at our center (1): prophylactic PLTX if morning PLC < 10 × 10⁹/L and (2) PLTX restricted to signs of bleeding as recently described (21), based on the treating physician’s discretion. Because minor bleeding signs were not always documented in the EMR, a prophylactic versus therapeutic approach could not assigned to every provided PLTX and was therefore not further integrated into analyses.

There were 1 to 2 units of packed red blood cell (PRBC) transfusions that were usually provided in case of hemoglobin ≤ 70 or 80 g/L with symptomatic anemia.

Categorical variables were summarized by number and percentage, and continuous variables were summarized by mean, standard deviation, and range (minimum/maximum). Continuous variables were compared between ASA and the control group (or as stated otherwise) using Welch’s t-test for unequal variances and categorial variables by the chi-square test. Univariate logistic regression models were built to test the association between BLEDs or PLTX requirements, respectively, and all relevant patient characteristics, clinical, and laboratory variables to determine the potential predictors of BLEDs or PLTX requirements. For PLTX requirements, patients with BLEDs were excluded from the analysis to avoid bias due to increased PLTX requirements during bleeding. Multivariable logistic regression models using stepwise estimation by backward and forward selection were then generated from variables with p ≤ 0.05 from the univariate models and known/expected factors to be associated with BLEDs or PLTX requirements. Firth’s logistic regression was used to confirm results for BLEDs as logistic regression could be biased by unbalanced predictors.

A linear regression model to assess risk factors for a prolonged duration of thrombocytopenia in patients without BLEDs to avoid a bias for increased transfusions during bleeding was built. Additionally, two outlier cases with an unusual requirement were excluded to avoid a bias for this analysis.

The rationale and sample size calculation for a randomized controlled trial to evaluate different ASA discontinuation strategies are described in the Supplement Material (1.1 Supplementary Methods , 1.1.2 Design of a randomized controlled trial to assess different APT strategies, and Supplementary Figure S1 ).

All statistical analyses were performed in SPSS version 28.0.0.0 and SAS version 9.4. Figures were created with R-studio version 4.1.2, Excel, and Biorender.

There were 13 patients who were excluded as they paused ASA intake earlier than 1 day after stem-cell reinfusion (day +1); therefore, neither ASA efficacy could be expected nor excluded (the patient characteristics of this group are depicted in Supplementary Table S1 ). Thus, 1,113 patients were included into analysis. Baseline patient characteristics are summarized in Table 1 . The patient characteristics of those who took ASA in comparison to those without (control group) revealed significant differences in age, gender, and the hematopoietic stem-cell transplantation comorbidity index (HCT-CI). Furthermore, patients in the ASA group had more CV comorbidities as expected. All other characteristics were similar between groups.

For 57 patients, continuous ASA efficacy during thrombocytopenia was assumed as ASA was continued until at least 1 day after ASCT. Of those, the majority (41/57) continued ASA until PLC 20–50 × 10⁹/L (11/57 patients stopped with PLC < 70–80 × 10⁹/L, 28/57 with PLC < 50 × 10⁹/L, 13/57 with PLC < 20–30 × 10⁹/L, and 2 with PLC < 10 × 10⁹/L. Only one patient continued ASA throughout thrombocytopenia, and for 2/57 patients, no clear strategy could be defined). The mean day of last ASA intake after ASCT was day +4.89 ± 1.37 (range: day 2–7), the mean days of ASA discontinuation were 9.39 ± 6.02 (range: 0–42 days; 95% confidence interval (CI) [8, 11.19]).

A clinically meaningful, documented BLED occurred in 23 patients (2.1%). Of these 23 patients, three patients were in the ASA group. Two of them discontinued ASA before the onset of a BLED, although a continued efficacy of APT was suspected. The third patient took ASA on the day of bleeding, but PLC was still normal. The clinical course of these three patients is depicted in Figure 1 . No significant difference in bleeding incidence was demonstrated for patients receiving ASA, although a higher tendency was observed [1.9% (control group) vs. 5.3% (ASA), p = 0.082].

The mean PLC at bleeding onset was 30.87 × 10⁹/L ± 46.48 (range: 4–179 × 10⁹/L). There were 52.2% (N = 12) of BLEDs that occurred during PLC ≤ 10 × 10⁹/L. In total, 82.6% of BLEDs were documented during PLC ≤ 50 × 10⁹/L ( Figure 2C ). The median day of bleeding onset was day +6 after ASCT (range: day −2 to +40). There were 16/23 BLEDs (69.6%) that occurred during a documented infection. The majority constituted lower GI bleedings (34.8%) ( Figure 2A ).

A specialist consultation (e.g., neurology and urology) was initiated in 8/23 patients, an endoscopy without further intervention was performed in two, and three patients underwent nasal packing to control epistaxis, which was the only intervention that was performed. The transfusion requirements and gradings of bleedings according to the WHO bleeding scale, CTCAE grading, and the BSMS are depicted in Figures 2B, D . In 16 patients, the BLED was deemed clinically significant according to the BSMS (comprising grade 2a/b). No relevant sequelae after hospital discharge were documented.

Of note, among patients who were excluded from analysis as they paused ASA intake earlier than 1 day after stem-cell reinfusion (day +1), therefore, no ASA efficacy was expected, one more patient developed a retinal hemorrhage. Thus, all three documented retinal hemorrhages occurred in patients with ASA. Further details on BLEDs are summarized in Supplementary Table S2 .

We further assessed the risk factors for BLEDs. Hypertension, the duration of PLC ≤ 50 × 10⁹/L, diarrhea, a history of GI bleeding, and the HCT-CI score were significantly (but weakly) associated with BLEDs (data not shown). The odds ratio (OR) for ASA intake was 2.78 (95% CI, [0.83, 9.98], p = 0.096). In multivariate binomial logistic regression, ASA intake as a risk factor for bleeding was also not significant ( Supplementary Figure S2A ). After multivariate binomial logistic regression by backward selection, only the duration of PLC ≤ 50 × 10⁹/L (OR = 1.166, 95% CI, [1.066, 1.277]; p < 0.001), diarrhea (OR = 4.038, 95% CI, [1.693, 9.632], p = 0.002), and a history of GI bleeding (OR = 24.230, 95% CI, [5.653, 103.861], p < 0.001) remained significantly associated with bleeding risk ( Figure 1E ). This binomial logistic regression model was statistically significant, χ² (3) = 31.500, p < 0.001. Goodness-of-fit as assessed using Hosmer–Lemeshow test indicated a good model fit, χ²(3) = 8.690, p > 0.050. The correlations between predictor variables were low (r < 0.70), indicating that multicollinearity was not a confounding factor in the analysis. As only few BLEDs occurred, results were confirmed with Firth regression to avoid biased results by disproportions among the groups. The complete model and Firth regression are depicted in Supplementary Figure S2A/B . The duration of PLC ≤ 10 × 10⁹/L was not associated with an increased bleeding risk ( Supplementary Figure S2C ).

Basic transfusion requirements and PLCs are summarized in Table 2 . A total of 477 (42.9%) patients received no PLTX at all; 481 (43.2%) received only one unit. Patients with ASA intake received significantly more units of PLTX (1.32 ± 1.42 vs. 0.78 ± 0.96, p < 0.001). No significant differences regarding the units of PRBCs or the duration of thrombocytopenia was demonstrated. The correlation between PLTX requirements and clinical factors is shown in Figure 3A . In binomial logistic regression analysis for PLTX requirements above the average of 0–1 unit of platelets, smoking, PRBCs, and the duration of PLC < 50 × 10⁹/L remained significant. ORs are depicted in Figure 3B . The binomial logistic regression model was statistically significant, χ²(15) = 198.747, p < 0.001. Goodness-of-fit as assessed by the Hosmer–Lemeshow test indicated a good model fit, χ²(15) = 7.326, p > 0.050. The correlations between predictor variables were low (r < 0.70). Since the duration of PLC < 50 × 10⁹/L showed a strong and clinical meaningful impact on PLTX requirements and BLEDs, risk factors for prolonged duration were further explored. The duration of PLC < 50 × 10⁹/L correlated significantly with age, PLC, hemoglobin, and prothrombin time (PTT) at admission, HCT-CI, hypertension, diarrhea, smoking, diabetes, hyperlipidemia, and infections ( Figure 3A ). The same factors were significant in univariant linear regression analysis. Based on that, a multiple linear regression model was built including all of these factors known before the start of conditioning. Smoking, diabetes, hypertension, and hyperlipidemia did not remain significant; thus, the strongest model was built including PLC at admission < 100 × 10⁹/L (ß = 0.241, p < 0.001), hemoglobin at admission < 100 g/L (ß = 0.073, p = 0.013), PTT at admission < 70% (ß = 0.103, p > 0.001), age > 60 years (ß = 0.111, p < 0.001), and HCT-CI ≥3 (ß = 0.084, p = 0.004). The R² for the overall model was 0.113 (adjusted R² = 0.109), indicating a weak goodness-of-fit according to Cohen. The model showed no autocorrelation as the value of the Durbin–Watson statistic was 1.698. An increased risk for BLEDs by a combination of these factors including PLC at admission < 100 × 10⁹/L, hemoglobin at admission < 100 g/L, PTT at admission < 70%, age > 60 years, and HCT-CI ≥3 was excluded by another binomial regression analysis using a sum score of these factors. The OR for this sum score was 1.194 (95% CI, [0.731, 19.52], p = 0.479).

Given the high frequency of the CV risk factor within our cohort, we assessed the incidence of MACE during the hospital stay. Three patients suffered a MACE. Two MACEs were likely associated with the transfusion of stem cells or a serious infection and are therefore discussed in the Supplementary Material (1.2 Supplementary Results and Discussion , 1.2.1 Case discussions on patients presenting with MACE, and Supplementary Table S3 . Cardiac comorbidities and risk factors in patients below versus above 60 years of age). Nonetheless, one patient was found unresponsive due to ventricular fibrillation and was successfully resuscitated. Subsequent coronary angiography revealed a significant occlusion of the left anterior descending artery (LAD), global hypokinesia, and reduced left ventricular ejection fraction. A drug-eluting stent and a single-chamber implantable cardiac defibrillator were implanted. The patient recovered with no obvious deficit. As CV risk factors, type II diabetes and hypertension were known prior to this event.

ASA represents the gold standard for preventing MACE (5). Despite its net benefit, major BLEDs were increased in both primary and secondary prevention trials (RR 2.69 [99% CI 1.25–5, p = 0.01) (5). Several retrospective analyses of cancer patients with thrombocytopenia and acute coronary syndrome suggested ASA and even dual APT to be safe and advantageous in regard to cardiac outcomes with a little risk of major bleeding (4, 27–30).

In our cohort, ASA was discontinued during thrombocytopenia of various degrees after high-dose chemotherapy. During aplasia, no or only a small percentage of new platelets are generated. Therefore, the remaining platelets were assumed to be continuously inhibited by ASA. The in vitro reversal of platelet inhibition in patients taking ASA was shown to be achieved by mixing inhibited platelets with 30% untreated donor platelets (31). The restoration of platelet function is mediated by TXA2 from (donor) platelets, which activates the thromboxane receptor of ASA-inhibited (host) platelets (31). This concept was supported by small case series and proof-of-principal studies in vivo (32–34). Despite, clinical outcome data on PLTX in patients taking ASA are scarce and the data from thrombocytopenic patients are not available at all. In our cohort, >70% of cases with ASA intake received only 0–1 unit PLTX after ASA pause during aplasia; thus, a relevant impact on continued ASA efficacy seems unlikely. Of note, if PLTX is provided during continued ASA intake, the strongest platelet inhibition by ASA 60 min after oral intake (35) should be considered, depending on the pursued effect.

We observed no MACE during the discontinuation of ASA in our cohort. Large analyses that assessed the discontinuation of ASA without surgery or bleeding after long-term use for secondary prevention demonstrated a 46% higher rate of MACE (7). This translates into one additional event in 1 of every 36 patients (7). Of note, the time from discontinuation to the thrombotic event was as early as 7–30 days in most reports (8–12, 36). In our cohort, patients discontinued ASA on average for ~9 days during thrombocytopenia. Although infections are related to a prothrombotic state, this time frame might be short enough to avoid a clinically relevant increased risk of MACE. In addition, the protective impact of thrombocytopenia for MACE is difficult to estimate or measure. This is supported by the fact that neither patients who discontinued ASA before day +1 nor patients who had an indication for secondary ASA intake but did not receive it experienced an additional MACE during ASCT. The latter group included notably 2.8% (N = 30) of control cases who had an indication for ASA as secondary prophylaxis and 0.3% (N = 3) for primary prophylaxis but did not receive it. Whether these patients developed an event after discharge could not be assessed as many of those received their follow-up visits outside our center. The only patient who developed a severe MACE during ASCT had no indication for preventive ASA to our knowledge based on his chart review.

Over 80% of BLEDs occurred during PLC < 50 × 10⁹/L and majority of these with a PLC < 10 × 10⁹/L. The severity of thrombocytopenia was repetitively shown not to correlate with BLEDs (38), although a PLC < 5 × 10⁹/L was demonstrated as a risk factor for BLEDs during ASCT (14). In our cohort, multivariate analysis validated the duration of PLC < 50 × 10⁹/L, not below < 10 × 10⁹/L, as a risk factor for bleeding. Although the duration of PLC < 10 × 10⁹/L might be skewed by PLTX, these results are in concordance with previous reports.

Three risk factors for bleeding remained statistically significant in multivariate analysis: a history of GI bleeding, diarrhea, and the duration of PLC < 50 × 10⁹/L. The number of patients with former GI bleeding was very low; thus, its predictive value might be overestimated. We therefore suggest to rather focus on factors that predict a prolonged duration of PLC < 50 × 10⁹/L than the direct risk factors for bleeding. Herein, a decreased bone marrow reserve (hemoglobin at admission < 100 g/L, PLC < 100 × 10⁹/L), impaired coagulation (PTT < 70% at admission), age >60 years, or HCT-CI ≥3 points are known prior to conditioning and can help to further guide physicians in assessing the indirect risk for bleeding. None of them were significantly associated with BLEDs itself in multivariate analysis ( Supplementary Figure S2A ), neither alone or as a sum score.

The recently reported risk factors for BLEDs comprised fever (HR: 1.7, 95% CI [1.3, 2.4]) (39, 40). Neither fever nor sepsis remained an independent risk factor for bleeding in the multivariate analysis of our cohort. This might be explained by the high number of infections during ASCT: 89.9% of cases developed at least one documented episode of fever/infection, and 69.6% of BLEDs occurred during an infection. Whether this constitutes a co-occurrence or a causality remains unclear, but a statistical association could not be demonstrated.

The incidence of BLEDs in our cohort was lower than recently reported (20, 21). In these trials, a daily structured bleeding assessment was mandatory and was documented as patient days with bleeding to account for the course of the bleeding with regard to PLTX. We documented every BLED only once; thereby, the bleeding incidence is hardly comparable. In addition, our data are based on EMRs without these structured assessments and the underreporting of mild BLEDs (e.g., mild epistaxis) is possible.

Furthermore, the number of patients on ASA was limited, although this is a large real-world cohort and the largest sample size assessing BLEDs during ASCT so far. Therefore, our results are hypothesis-generating and this question can only be answered in a randomized controlled trial. Such a trial would require a large international cooperative initiative that does not appear to be feasible (for the design and sample size calculation of such a trial, please refer to Supplementary Material , 1.1 Supplementary Methods , 1.1.2 Design of a randomized controlled trial to assess different APT strategies, and Supplementary Figure S1 ).

As an evidence-based strategy guided by high-quality randomized trial data will most likely never be available, the following rationale-based approach can be considered: if ASA is indicated based on a primary preventive approach, the discontinuation of ASA before conditioning can be considered based on the high number needed to treat (NNT) to prevent MACE and the relevant risk for bleeding even with normal PLC. If ASA is indicated by a secondary preventive approach, it appears safe to continue ASA until a PLC of 20–50 × 10⁹/L (CTCAE III°). We consider this PLC range rather than a fixed cut-off to pause ASA, as PLC was not measured daily in our routine clinical practice. Furthermore, a PLC range of 20–50 × 10⁹/L reflects the real-world scenario better than a fixed cut-off given the standard kinetics of thrombocytopenia during ASCT. Whether the application of ASA even during PLC < 20 × 10⁹/L with a lower transfusion trigger (e.g., 20–30 × 10⁹/L) is also safe cannot be concluded from our data as only one patient was guided with this strategy. The rationale behind such an approach is a potentially increased prothrombotic state due to a rebound effect after the discontinuation of ASA. When the ASA discontinuation strategy for the individual patient is discussed, we recommend to consider age >60 years, HCT-CI ≥3, PLC < 100 × 10⁹/L, hemoglobin < 100 g/L, and PTT < 70% at admission as predictors for a prolonged duration of PLC < 50 × 10⁹/L, and therefore, an indirect bleeding risk together with a clinical bleeding history (especially GI bleedings). Recommendations are summarized in Figure 4 ; general results are shown in Figure 5 .