Serial Platelet Level Index Improves Prediction of Pulmonary Hemorrhage in Stenotrophomonas maltophilia Respiratory Infections

Background: Thrombocytopenic patients who acquire Stenotrophomonas maltophilia (SM) respiratory infection often develop pulmonary hemorrhage (PH), resulting in a high respiratory failure rate and increased mortality. This study aimed to evaluate risk factors for PH and develop an index measuring serial platelet decit to predict PH in patients with SM respiratory infection. Methods: Data of patients with both SM isolated from sputum/endotracheal suction culture and thrombocytopenia (platelet count < 150x10 3 /μL) who were treated at National Cheng Kung University Hospital during 2018-2020 were extracted from electronic medical records and analyzed retrospectively. SM respiratory infection was dened as positive bacterial isolates plus respiratory infection symptoms. Clinical parameters and laboratory ndings were compared between PH and non-PH groups. The platelet dissimilarity index (d-index) was calculated by accumulating differences between the actual and the lowest normal level of platelet count in each patient at different time points. Results: Among 437 SM respiratory infection cases, 125 (28.6%) patients developed PH. Patients with PH had increased prothrombin time/international normalized ratio (PT/INR), lower platelet count, and higher platelet d-index. Multivariate analysis revealed that extreme thrombocytopenia (platelet count < 50x103 /μL) is a common independent risk factor for PH and mortality. The performance of platelet decit and d-index varied between patients with different comorbidities. Performance of single time-point platelet decit to predict PH is more reliable in patients with hematology/oncology and liver disease (area under curve 0.705-0.757), while performance of d-index is more reliable in patients with sepsis/treatment and various other groups (0.711-0.816). Conclusions: Prolonged and extreme thrombocytopenia is a determinant risk factor for PH in patients with SM respiratory infection. Given the complexity of causes of thrombocytopenia and associated comorbidities, different strategies should be applied to assess the degree of thrombocytopenia when evaluating the risk for PH. To assess the ability of platelet d-index for different time windows and platelet decits of a single day to predict pulmonary hemorrhage, sensitivity and specicity were analyzed using receiver operating characteristic (ROC) curves across the sputum sampling day showing SM positive. The optimal cut-off value was determined by the maximum sum of sensitivity and specicity.

("Liver"); and various other causes ("Others") (e.g., mechanical destruction, transfusion consumption coagulopathy or major bleeding, unknown cause). PH was de ned in patients with multiple episodes of hemoptysis and desaturation or evidence of deteriorating pulmonary function or anatomy (e.g., increased in ltrations in chest X-ray, increased oxygen demand).
Platelet D-Index and Platelet De cit Calculation D-index was applied in patients with prolonged neutropenia and opportunistic infections using an accumulated difference between neutrophil count and the threshold of neutropenia, as described previously [12][13][14]. The same rationale was applied to estimate the accumulated platelet de cit in thrombocytopenic patients. The Platelet D-index was calculated using this algorithm.
PlateletD index = A e − A 0 = ∑ n i = 2 150 * t i − t i − 1 − ∑ n i = 2 P i − 1 + P i 2 * t i − t i − 1 )] Accumulated platelet de cit between estimated area under curve (Ae) and observed area under curve (Ao) 7 days before and after sputum culture result of SM was calculated. Time window of platelet d-index included 3 days before and after SM sputum test date (d-index-6), 5 days before and after (d-index-10), and 7 days before and after (d-index-14). The platelet de cit was calculated using the lowest level of normal platelet count (150 x10 3 /µL) minus the platelet count at test date.

Statistical Analysis
All statistical analyses were performed using SPSS 17 for Windows (IBM SPSS, Armonk, NY, USA). Patients with SM respiratory infections were grouped as "with PH" or "without PH". The independent sample t-test was used for continuous variables, and the Chi-square test or Fisher's exact test was used for dichotomous variables. P value < 0.05 was considered statistically signi cant. Cox's proportional hazard analysis was used to determine risk factors for PH and 30-day mortality. One-way ANOVA was used to assess platelet levels and d-index of various causes of thrombocytopenia. To assess the ability of platelet d-index for different time windows and platelet de cits of a single day to predict pulmonary hemorrhage, sensitivity and speci city were analyzed using receiver operating characteristic (ROC) curves across the sputum sampling day showing SM positive. The optimal cut-off value was determined by the maximum sum of sensitivity and speci city.

Study population
Of the 437 patients with thrombocytopenic SM respiratory infection, 125 patients (28.6%) developed PH. Patients' demographic and clinical characteristics are shown in Table 1. No signi cant differences were found in age, sex, and cause of thrombocytopenia between patients with and without PH. The most common causes of thrombocytopenia were "Sepsis-Rx" (57.0%), followed by "Others" (15.1%), "Liver disease" (14.9%), and "Hem-Onc" (13.0%). Although hematological disease accounts for the majority (48/57, 84.2%) of cases in the "Hem-Onc" group, the proportion of hematological diseases was not signi cantly different between the "Hem-Onc and "Liver" groups. Platelet counts at the test date were lower, and the d-index of different time frames was higher in the PH group than in the non-PH group (P value < 0.001). Also, prolonged PT/INR and mechanical ventilation use were signi cantly associated with PH (all P value < 0.05). The mortality of SM respiratory infection was higher in the PH group than in the non-PH group (87.2% vs. 57.7%, P value < 0.001), with the overall rate of 66.1%. Half of the patients with PH died within the rst week after diagnosis, and the mortality rate declined in the second and third weeks.
None of the neutropenic patients with PH (n = 15) survived.
[ ( )] [ ( Risk factors associated with pulmonary hemorrhage and mortality Table 2 and Supplementary Table 2 list the results of Cox proportional hazard analysis of risk factors associated with PH in neutropenic SM respiratory infections and risk factors associated with 30-day mortality in the PH group, respectively. In univariate analysis, neutropenia, and extreme thrombocytopenia (platelet count at test date < 50 x10 3 /µL) were both risk factors for PH and 30-day mortality. However, multivariate analysis showed that severe thrombocytopenia was the only independent risk factor for both PH and 30-day mortality in PH patients.

Characteristics of SM respiratory infections by cause of thrombocytopenia
Comparison of clinical characteristics and laboratory ndings between patients with thrombocytopenic SM respiratory infection caused by thrombocytopenia is shown in Table 3. The "Sepsis-Rx" group had more patients aged 65 years and older (67.9%), higher pulmonary structural abnormalities (41.8%), and greater need for mechanical ventilation (86.7%). Platelet counts at the test date were signi cantly lower in patients in "Hem-Onc" and "Liver" groups than those in the "Sepsis-Rx" and "Others" groups, with the most extreme thrombocytopenia (platelets < 50 x10 3 /µL) cases in the "Hem-Onc" group. The d-index-14 was inversely correlated with platelet counts at the test date and was signi cantly higher in patients in the "Hem-Onc" and "Liver" groups than those in the "Sepsis-Rx" and "Others" groups. However, the percentage of PH and the time to developing PH from diagnosis were not signi cantly different between these four groups. Mortality rates (66.1%) were signi cantly higher in the "Hem-Onc" (71.9%) and "Sepsis-Rx" (71.9%) groups than those in the "Liver" (52.3%) and "Others" groups (53%).  Fig. 1 and Table 4. The predictability of these parameters varied across different causes of thrombocytopenia, and d-index-10 had the highest AUC (0.719) in all cases. Platelet de cit performed more reliably than d-index in the "Hem-Onc," and "Liver" groups, and the two groups combined. In contrast, d-index performed more reliably in the "Sepsis-Rx" and "Others" groups. D-index-14 of the "Others" group showed the highest rate of predictability (AUC = 0.816) of PH across all groups. The cutoff values of platelet de cit and d-index-14 according to the maximum sums of sensitivity and speci city are listed in Table 4.
The optimal cutoff value of platelet de cit on the test date for all cases was 70.5 x10 3 /uL, corresponding to platelet levels of 79.5 x10 3 /uL with a sensitivity of 69.8% and a speci city of 57.7% in predicting PH. The optimal platelet cut-off value at the test date varied across the four groups, from 17.5 x10 3 /uL in the "Hem-Onc" group, to 67.5 x10 3 /uL in the "Liver" group, 86.5 x10 3 /uL in the "Sepsis-Rx" group, and 106 x10 3 /uL in the "Others" group. We subtracted the dindex-14 by days to calculate the average platelet de cit per day and the corresponding value had a similar pattern as the platelet de cit at the test date, but with more reliability in "Sepsis-Rx" and "Others" group. AUC, area under curve; cutoff value (/µL) indicates a value that could with the optimal sum of highest sensitivity and speci city. platelet de cit on the test date indicated the de cit of platelet count from lowest normal limit on the test date (150 x103/µL -platelet count).

Discussion
Although thrombocytopenia had been recognized as a signi cant risk factor for PH in SM respiratory infection, in the present study, the degree of thrombocytopenia by platelet de cit at a single time point and over a certain time period was evaluated to identify the optimal cut-off value in predicting PH.
Also, results of the present study demonstrated the clinical heterogeneity of thrombocytopenia as the cause and its association with PH, which may explain the con icting ndings of previous observational studies and address the need for more precise and individualized risk classi cation based on comorbidities.
Different measurement tools should be used in distinct subpopulations to predict PH; for example, platelet counts at the test date should be su cient in patients with hematological and liver diseases, while d-index using serial platelet measurements performs more reliably for those with sepsis and other causes.
Mortality associated with PH was overwhelmingly high in the present study, especially in patients with hematologic malignancy. Recently, increasing numbers of cases have been successfully treated, which may be due to re nement of preventive measures of transmission, rising awareness of SM infections, and timely use of empirical antimicrobial therapy [18,19]. The mortality rate in the present study was highest within the rst week of diagnosis (50%), demonstrating the potential fulminant behavior of SM infections. Notably, only one patient with hematological disease survived through hemorrhagic SM respiratory infection, and none of the neutropenic patients survived. In a previous study conducted by Bao et al, patients with severe prolonged neutropenia and thrombopenia due to hematological disorders were shown to have 100% mortality when encountering SM bacteremia [20]. The present study echoes their ndings, indicating that hematological abnormalities complicate the clinical course of SM respiratory infection and worsen the disease outcome. Although neutrophil counts may have been a decisive factor in infection control and hematopoiesis ability, it was also shown to be a signi cant risk factor for PH in previous studies [15]. After multivariate analysis, we found that neutropenia was not an independent risk factor for PH or associated mortality, implying that neutropenia is probably a collateral nding along with underlying disease and thrombocytopenia.
Patients in the "Sepsis-Rx" and "Others" groups with SM-associated PH had higher proportions of pulmonary structural abnormalities and mechanical use.
Because these patients are likely to receive mechanical ventilation due to pulmonary structure anomaly, they acquired SM infection through ventilator use.
Until now, the attributable factors causing SM-related VAP remain unclear [1,16]. A recent study found that exposure to ureido/carboxypenicillin or carbapenem during the week before VAP, and the severity of disease leading to respiratory and hematological failures, were independent risk factors for SM-VAP [1]. Moreover, SM-VAP mortality remains high even in patients receiving adequate treatment, either monotherapy or combinations of antimicrobials.
Platelet counts at a single time point or over a given time period were signi cantly lower in patients with hematological and liver diseases than in those with sepsis or other diseases, but risk for PH did not differ signi cantly between these patients. Although thrombocytopenia is an independent determining factor for PH, patients experiencing long-term thrombocytopenia may develop a certain mechanism by which to compensate for the bleeding tendency. A previous study identi ed that an increased number of larger-sized platelets may compensate for the impaired platelet function in patients with chronic idiopathic thrombocytopenia [17]. In acute myeloid leukemia patients who have thrombocytopenia, platelet aggregation and platelet activation predicted bleeding better than the platelet count alone [18]. Overall, the bleeding risk is not only dependent on the platelet count but also on the platelet function, coagulopathy, and the underlying disease causing thrombocytopenia. Further mechanistic exploration of the cause of reduced platelet counts or function in different thrombocytopenic conditions is still needed to help develop and implement preventive strategies [19].
In the present study, thrombocytopenia due to hematological and liver diseases both had prolonged and profound low platelet levels and PH usually developed at the nadir of the platelet level. Meanwhile, the pattern of thrombocytopenia due to sepsis/medication or other causes were more alike and had higher average platelet count compared to those in hematological and liver disease. Traditionally, the pathophysiology of thrombocytopenia in chronic liver disease has long been attributed to hypersplenism, where pooling and sequestration of blood results in platelet consumption. Recently, other mechanisms, including bone marrow suppression by toxic substances such as alcohol or viral infection may also contribute to thrombocytopenia. In addition, the thrombopoietin, predominantly produced by the liver is markedly reduced in advanced-staged liver disease, which also contributes to reduced thrombopoiesis in the bone marrow [20]. All mechanisms mentioned above can explain the thrombocytopenia caused by liver disease that may be secondary to bone marrow suppression, similar to those in hematological disorders.
The clinical consensus regarding the lowest platelet level needed to prevent bleeding in certain circumstances is that platelet counts must be above 20-50 x10 3 /uL for bronchoscopy exams and 50 x10 3 /µL for transbronchial lung biopsies [21]. Statistical results from the present study show that platelet levels above 60-100 x10 3 /µL may plausibly prevent PH in patients with sepsis, liver disease and other causes, while lower platelet counts (above 17.5 x10 3 /µL) may be tolerated in patients with hematological diseases. In addition to certain cut-off levels for platelet counts, the accumulative platelet de cit over a given time period was found to be more reliable in predicting PH in patients with sepsis, medication-related, or other causes of thrombocytopenia. Although these groups of patients did not have thrombocytopenia as profound as in those with hematological disorders, they are more susceptible to developing PH in prolonged thrombocytopenia. Therefore, regular prophylactic platelet transfusion to maintain platelet levels above certain thresholds over unstable periods may be a potential strategy to prevent PH in this group of patients.
The present study used the largest database evaluating PH in patients with SM respiratory infections and is also the rst study to extrapolate serial platelet measurements in predicting PH. Data in the present study provide clinicians with the optimal cutoff platelet level for transfusion therapy to prevent PH in thrombocytopenia of various causes. The concept of using d-index as a personalized measurement in evaluating the severity of thrombocytopenia can be implemented to prevent SM hemorrhagic pneumonia and bleeding disorders in various other diseases. Especially in the era of the COVID-19 pandemic, clinicians may also utilize the d-index to assess the risk of PH in sepsis due to COVID-19.
Regardless of the above strengths, the present study has several limitations. First is that we did not evaluate associations between PH, antibiotic use and corresponding antimicrobial susceptibility. Although in both clinical and animal studies, timely appropriate antibiotic use could reduce the risk of mortality in SM hemorrhagic pneumonia, systematic review demonstrated that even patients treated with appropriate antibiotics, including trimethoprim/sulfamethoxazole, uoroquinolones, or both combined, the mortality remained high, even reaching 100% [11,15]. Also, no prior antimicrobial therapy for SM bacteremia had shown a preventive effect for PH [7]. Still, prompt administration of antibiotics is essential for control of infectious disease, but the use of antibiotics in preventing PH in SM respiratory infection does not appear to be an adequate measure.

Conclusion
Prolonged and extreme thrombocytopenia is a determinant risk factor for PH in patients with SM respiratory infections. The degree of platelet de cit varies signi cantly between different causes of thrombocytopenia and types of underlying comorbidities. Single time-point measurement of platelet counts may most reliably predict PH in patients with hematological and liver diseases, while serial measurement over a given time period may reliably predict PH in those with sepsis and other causes of thrombocytopenia.