Introduction
Patients with acute leukemia frequently suffer from bleeding events [
1] of which intracranial hemorrhage (ICH) is one of the most serious [
2‐
5]. Reported incidences of (symptomatic) intracranial hemorrhage vary between 2.8% up to 6.1% [
2,
5,
6], and fatal intracranial hemorrhages explain more than 50% of fatal bleedings among acute leukemia patients [
7].
Acute leukemia patients may develop intracranial hemorrhage due to various causes. Besides risk factors that also play a role in the general population, like age, hypertension, male sex, and ethnicity [
8‐
10], leukemia or cancer-specific risk factors have been established. Among others, these are graft versus host disease, hyperleukocytosis, and thrombocytopenia [
11‐
14]. Of these, the low platelet count is generally considered one of the most important risk factors for bleeding in hemato-oncological patients. It is, however, not conclusively established if, and at what platelet counts, the risk of intracranial hemorrhage increases in this patient population [
2,
5‐
7,
12,
15,
16]. Moreover, prolonged exposure to low platelet counts (≤ 10 × 10
9/L) may be associated with even higher bleeding risks [
17,
18]. We hypothesized that longer periods with low platelet counts as well as lower (through) platelet counts can both determine an increasing risk of intracranial hemorrhage. If these time and trough measures are stronger associated with bleeding risk, this could have implications for future treatment strategies.
To prevent bleeding, hemato-oncology patients with low platelet counts are generally treated with prophylactic platelet transfusions [
19‐
21]. The trigger to transfuse is commonly set at a platelet count of 10 × 10
9/L [
22‐
24]. Prophylactic platelet transfusions reduced the risk of bleedings in patients with a World Health Organization (WHO) score of ≥ 2 [
25] from 50 to 43% [
26], with the most benefit for patients with acute myeloid leukemia or intensive chemotherapy treatment [
16,
17,
27]. However, the large majority of bleeds is thus not prevented despite platelet transfusions. This raises questions about the causes of bleeding both when patients are treated with prophylactic platelet transfusions and also when they are not. Interestingly, recent high-level evidence suggests that among neonates and among patients with hemorrhagic stroke, both prophylactic and therapeutic platelet transfusions may increase the risk of bleeding and/or mortality and morbidity [
28,
29].
How exactly the depth and length of thrombocytopenia and the given platelet transfusions interact and modulate the risk of critical bleeding like intracranial hemorrhage is presently unknown.
Therefore, the objective of this exploratory study was to describe the association of platelet counts assessed in several time periods and severities with the incidence of intracranial hemorrhage in acute leukemia patients. Also, we wanted to examine the association between platelet transfusions and the incidence of intracranial hemorrhage.
Discussion
In this case-control study among leukemia patients, we observed that one or more platelet counts below thresholds of both 10 × 109/L and 20 × 109/L, and an increasing percentage of hours below 20 × 109/L was associated with intracranial hemorrhage, especially when low platelet counts occurred more than one day before the event of the hemorrhage. However, the estimates of these associations lacked precision. Platelet transfusions were also associated with the occurrence of subsequent intracranial hemorrhage; these estimates of association were likewise imprecise.
The point estimates of the association between all the defined measures of low platelet counts and the incidence of intracranial hemorrhage show a clear trend of higher incidences of intracranial hemorrhage when platelet counts are low. The most likely rate ratios are especially increased if platelet counts were low at three, five or seven days before the hemorrhage. In contrast, no increased incidence is seen in the period of 1 day before hemorrhage for two out of our three defined measures of platelet count. Although almost all point estimates go in the same direction, and an increased incidence of intracranial hemorrhage when platelet counts are low is thus most likely, the confidence intervals are wide, due to low numbers of patients. This means that the true effect size could lay in a wide range of values, from strongly harmful to even protective.
Quantitative evidence on the association between platelet counts and the occurrence of intracranial hemorrhage among patients with leukemia is scarce. Some reports focused on fatal intracranial hemorrhage [
2,
5‐
7,
15]. One study did find an association between thrombocytopenia and the occurrence of intracranial hemorrhage in a subgroup of post-allogeneic stem cell transplantation patients [
12]. Two RCTs investigated the effect of prophylactic platelet transfusions on the occurrence of bleeding. Therapeutically treated patients had lower platelet counts compared with prophylactically transfused patients. One RCT did not find a difference in the occurrence of grade 3 and 4 bleedings (including intracranial hemorrhage) [
26] while the other did see more intracerebral hemorrhage in the therapeutically transfused group [
16]. However, the latter RCT had a different CT scan policy for both study arms, which likely reduced the number of confirmed intracranial hemorrhage in the control arm.
Moreover, most studies describe associations of bleeding with platelet counts of only one day or do not clarify fully which platelet counts are taken into consideration for the analysis. However, it has also been suggested that there may be a longer lag time before low platelet counts can lead to bleeding [
18]. Our results suggest that potentially prolonged thrombocytopenia (3 to 7 days) is leading to more intracranial hemorrhages. Our study is as far as we know the first to define several implicated periods and several measures of platelet count, to investigate the association between both time and trough of low platelet counts and intracranial hemorrhage.
Platelet counts are not surprisingly strongly related to platelet transfusions in this patient population. Low platelet counts lead to transfusions, and transfusions affect future platelet counts. Since in this study we also saw an association between platelet transfusions and intracranial hemorrhage, ideally you would like to adjust for the potential confounding effect of platelet transfusions. However, this is extremely difficult, even if one would have a large dataset, given that platelet counts and platelet transfusions are so strongly interdependent, and that multiple platelet counts, and transfusions would need to be considered (see online supplementary material: fig.
S1). In our small sample size, such corrections are impossible.
In the present study, also platelet transfusions were associated with an increased incidence of intracranial hemorrhage, especially when > 2 transfusions were given in an implicated period.
Since low platelet counts are often the reason for platelet transfusion, we aimed to correct for the defined measures of platelet count. Due to the fact that patients often had multiple transfusions and multiple platelet count determinations, a reliable and complete correction is again not possible in our dataset. Nevertheless, by adding the different defined measures of platelet count into the model, we see that this did not influence the observed association between platelet transfusion and intracranial hemorrhage in our study. Therefore, we infer that it seems plausible that the need for platelet transfusions or platelet transfusions itself in the circumstances where they are frequently needed might increase the incidence of intracranial hemorrhage and that this is at least partly independent of platelet counts. However, other clinical factors that lead to an increasing need for platelet transfusions, for example, conditions leading to increased platelet consumption, are very likely responsible for the latter observed association with intracranial hemorrhage. To investigate the impact of such potential confounding clinical conditions, we corrected them by adding relevant clinical factors in the regression model. Indeed, we identified anticoagulation/antiplatelet therapy and other (non-intracranial) bleeding events as possible confounders. These were also previously suggested to increase bleeding risk in hemato-oncology patients [
22]. For causal interpretation, an extensive multivariable model in an individual patient data meta-analysis of studies like ours would be essential allowing adjustment for all confounding. Besides confounding, the observed association between platelet transfusions and intracranial hemorrhage may also be due to relative functional defects of the transfused platelets. Platelet concentrates are known to develop storage lesions, which can lead to reduced platelet quality [
33,
34]. Moreover, one could argue that the transfusions contribute to intracranial hemorrhage by other mechanisms. Platelets do not only act in primary hemostasis but also have immunomodulatory functions. Inflammation is likely to influence bleeding risks, especially in thrombocytopenic conditions [
17,
35‐
39]. The idea that platelet transfusions lead to adverse outcomes is indeed reported by two RCTs, both showing adverse effects on morbidity and mortality in very different patients’ populations, namely patients with a hemorrhagic cerebral vascular accident while using antiplatelet agents and thrombocytopenic neonates [
28,
29]. The mechanisms behind these findings, however, are unclear. Finally, the observed associations could also be due to chance.
Strengths and limitations
A strength of this study is the matching of case and control patients on diagnosis and treatment. This allowed adjustment for these important known risk factors for this rare, but feared, bleeding complication.
Also, matching on the hospital was performed to correct for confounders that are not easily quantified, like differences in local treatment policies. Additionally, we matched case and control patients on time after starting a treatment or after the admission date. During admission, a leukemia patient is likely exposed to different platelet counts and other clinical risk factors, mostly determined by the exposure to intensive cytoreductive treatment. By matching case and control patients on time after therapy/admission, we minimized confounding by direct treatment effects.
Another asset of the study is the completeness of information for our main variables, namely platelet counts and platelet transfusions. A strong feature of the study is that we examined multiple measures of platelet counts during a week before the intracranial hemorrhage. With these different measures, we could explore various possible influences of thrombocytopenia, like trough level and duration, on the incidence of intracranial hemorrhage during one, three, five and seven days before the hemorrhage. To the best of our knowledge, this has not been performed in other studies.
Finally, our study may be a novel framework which enables taking time aspects and thrombocytopenia severity into account. Our nested case-control study, that to our knowledge was not applied before, allowed exploration of effects of time and severity, via defining various implicated time periods for multiple measures of the exposure on the outcome intracranial hemorrhage.
Our study has also some limitations. First, our sample size was too small to assess some potentially interesting and relevant measures of platelet count. Since patients are transfused as soon as platelet counts drop below 10 × 109/L, the time below this value could not be sufficiently assessed. Even with a larger study population, the frequent transfusions would likely still minimize the amount of time ≤ 10 × 109/L. Therefore, although this cutoff point is the most widely used transfusion trigger, we could not assess the effect of time below 10 × 109/L on the occurrence of intracranial hemorrhage.
Furthermore, as discussed earlier, due to the small sample size, we could only correct for one variable at the time. Therefore, by the lack of multivariate analysis, residual confounding remains. While we aimed to assess causality, although proving causality is never possible [
40,
41], all results have to be interpreted as hypothesis generating only. Confirmation in larger studies will be necessary, although challenging due to the rarity of intracranial hemorrhage. In addition, biological mechanisms should be investigated.
Also, we may have missed patients that acutely died due to severe intracranial hemorrhage, leading to potential bias. These patients remain undetected in the applied algorithm due to the absence of laboratory or additional diagnostics. The number of these missed patients is likely to be very limited. So, a relevant change of the findings is not to be expected, except for inducing a lower incidence of intracranial hemorrhage. Finally, given the retrospective nature of collecting data, it was not always possible to distinguish if platelet transfusions were truly prophylactic. Transfusion triggers were often not recorded clearly and might have been higher than 10 × 10
9/L in case of an assumed higher bleeding risk [
22‐
24]. Possibly also some therapeutic transfusions might have been included if they were actually given for an unrecorded (probably minor) bleeding event. Patients who already need therapeutically platelet transfusions have proven to be more prone to bleeding and thereby are likely to also have a higher risk for intracranial bleeding.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.