Epigenetic control over cell-intrinsic immune response antagonizes self-renewal in acute myeloid leukemia.

Epigenetic modulation of the cell-intrinsic immune response holds promise as a therapeutic approach for leukemia. However, current strategies designed for transcriptional activation of endogenous transposons and subsequent interferon type-I (IFN-I) response, show limited clinical efficacy. Histone lysine methylation is an epigenetic signature in IFN-I response associated with suppression of IFN-I and IFN stimulated genes, suggesting histone demethylation as key mechanism of reactivation. In this study, we unveil the histone demethylase PHF8 as a direct initiator and regulator of cell-intrinsic immune response in acute myeloid leukemia (AML). Site-specific phosphorylation of PHF8 orchestrates epigenetic changes that upregulate cytosolic RNA sensors, particularly the TRIM25-RIG-I-IFIT5 axis, thereby triggering the cellular IFN-I response-differentiation-apoptosis network. This signaling cascade largely counteracts differentiation block and growth of human AML cells across various disease subtypes in vitro and in vivo. Through proteome analysis of over 200 primary AML bone marrow samples, we identify a distinct PHF8/IFN-I signature in half of the patient population, without significant associations with known clinically or genetically defined AML subgroups. This profile was absent in healthy CD34-positive hematopoietic progenitor cells, suggesting therapeutic applicability in a large fraction of AML patients. Pharmacological support of PHF8 phosphorylation significantly impairs growth of primary AML patient samples. These findings provide novel opportunities for harnessing the cell-intrinsic immune response in the development of immunotherapeutic strategies against AML.

Integration of Patient-Reported Outcome Data Collected Via Web Applications and Mobile Apps Into a Nation-Wide COVID-19 Research Platform Using Fast Healthcare Interoperability Resources: Development Study.

BACKGROUND: The Network University Medicine projects are an important part of the German COVID-19 research infrastructure. They comprise 2 subprojects: COVID-19 Data Exchange (CODEX) and Coordination on Mobile Pandemic Apps Best Practice and Solution Sharing (COMPASS). CODEX provides a centralized and secure data storage platform for research data, whereas in COMPASS, expert panels were gathered to develop a reference app framework for capturing patient-reported outcomes (PROs) that can be used by any researcher. OBJECTIVE: Our study aims to integrate the data collected with the COMPASS reference app framework into the central CODEX platform, so that they can be used by secondary researchers. Although both projects used the Fast Healthcare Interoperability Resources (FHIR) standard, it was not used in a way that data could be shared directly. Given the short time frame and the parallel developments within the CODEX platform, a pragmatic and robust solution for an interface component was required. METHODS: We have developed a means to facilitate and promote the use of the German Corona Consensus (GECCO) data set, a core data set for COVID-19 research in Germany. In this way, we ensured semantic interoperability for the app-collected PRO data with the COMPASS app. We also developed an interface component to sustain syntactic interoperability. RESULTS: The use of different FHIR types by the COMPASS reference app framework (the general-purpose FHIR Questionnaire) and the CODEX platform (eg, Patient, Condition, and Observation) was found to be the most significant obstacle. Therefore, we developed an interface component that realigns the Questionnaire items with the corresponding items in the GECCO data set and provides the correct resources for the CODEX platform. We extended the existing COMPASS questionnaire editor with an import function for GECCO items, which also tags them for the interface component. This ensures syntactic interoperability and eases the reuse of the GECCO data set for researchers. CONCLUSIONS: This paper shows how PRO data, which are collected across various studies conducted by different researchers, can be captured in a research-compatible way. This means that the data can be shared with a central research infrastructure and be reused by other researchers to gain more insights about COVID-19 and its sequelae.

Next-generation study databases require FAIR, EHR-integrated, and scalable Electronic Data Capture for medical documentation and decision support.

Structured patient data play a key role in all types of clinical research. They are often collected in study databases for research purposes. In order to describe characteristics of a next-generation study database and assess the feasibility of its implementation a proof-of-concept study in a German university hospital was performed. Key characteristics identified include FAIR access to electronic case report forms (eCRF), regulatory compliant Electronic Data Capture (EDC), an EDC with electronic health record (EHR) integration, scalable EDC for medical documentation, patient generated data, and clinical decision support. In a local case study, we then successfully implemented a next-generation study database for 19 EDC systems (n = 2217 patients) that linked to i.s.h.med (Oracle Cerner) with the local EDC system called OpenEDC. Desiderata of next-generation study databases for patient data were identified from ongoing local clinical study projects in 11 clinical departments at Heidelberg University Hospital, Germany, a major tertiary referral hospital. We compiled and analyzed feature and functionality requests submitted to the OpenEDC team between May 2021 and July 2023. Next-generation study databases are technically and clinically feasible. Further research is needed to evaluate if our approach is feasible in a multi-center setting as well.

Narcissistic dimensions and depressive symptoms in patients across mental disorders in cognitive behavioural therapy and in psychoanalytic interactional therapy in Germany: a prospective cohort study.

BACKGROUND: Narcissistic personality traits have been theorised to negatively affect depressive symptoms, therapeutic alliance, and treatment outcome, even in the absence of narcissistic personality disorder. We aimed to examine how the dimensional narcissistic facets of admiration and rivalry affect depressive symptoms across treatment modalities in two transdiagnostic samples. METHODS: We did a naturalistic, observational prospective cohort study in two independent adult samples in Germany: one sample pooled from an inpatient psychiatric clinic and an outpatient treatment service offering cognitive behavioural treatment (CBT), and one sample from an inpatient clinic providing psychoanalytic interactional therapy (PIT). Inpatients treated with CBT had an affective or psychotic disorder. For the other two sites, data from all service users were collected. We examined the effect of core narcissism and its facets admiration and rivalry, measured by Narcissistic Admiration and Rivalry Questionnaire-short version, on depressive symptoms, measured by Beck's Depression Inventory and Patient Health Questionnaire-Depression Scale, at baseline and after treatment in patients treated with CBT and PIT. Primary analyses were regression models, predicting baseline and post-treatment depression severity from core narcissism and its facets. Mediation analysis was done in the outpatient CBT group for the effect of the therapeutic alliance on the association between narcissism and depression severity after treatment. FINDINGS: The sample included 2371 patients (1423 [60·0%] female and 948 [40·0%] male; mean age 33·13 years [SD 13·19; range 18-81), with 517 inpatients and 1052 outpatients in the CBT group, and 802 inpatients in the PIT group. Ethnicity data were not collected. Mean treatment duration was 300 days (SD 319) for CBT and 67 days (SD 26) for PIT. Core narcissism did not predict depression severity before treatment in either group, but narcissistic rivalry was associated with higher depressive symptom load at baseline (ß 2·47 [95% CI 1·78 to 3·12] for CBT and 1·05 [0·54 to 1·55] for PIT) and narcissistic admiration showed the opposite effect (-2·02 [-2·62 to -1·41] for CBT and -0·64 [-1·11 to -0·17] for PIT). Poorer treatment response was predicted by core narcissism (ß 0·79 [0·10 to 1·47]) and narcissistic rivalry (0·89 [0·19 to 1·58]) in CBT, whereas admiration showed no effect. No effect of narcissism on treatment outcome was discernible in PIT. Therapeutic alliance mediated the effect of narcissism on post-treatment depression severity in the outpatient CBT sample. INTERPRETATION: As narcissism affects depression severity before and after treatment with CBT across psychiatric disorders, even in the absence of narcissistic personality disorder, the inclusion of dimensional assessments of narcissism should be considered in future research and clinical routines. The relevance of the therapeutic alliance and therapeutic strategy could be used to guide treatment approaches. FUNDING: IZKF Münster. TRANSLATION: For the German translation of the abstract see Supplementary Materials section.

Active learning for extracting surgomic features in robot-assisted minimally invasive esophagectomy: a prospective annotation study.

BACKGROUND: With Surgomics, we aim for personalized prediction of the patient's surgical outcome using machine-learning (ML) on multimodal intraoperative data to extract surgomic features as surgical process characteristics. As high-quality annotations by medical experts are crucial, but still a bottleneck, we prospectively investigate active learning (AL) to reduce annotation effort and present automatic recognition of surgomic features. METHODS: To establish a process for development of surgomic features, ten video-based features related to bleeding, as highly relevant intraoperative complication, were chosen. They comprise the amount of blood and smoke in the surgical field, six instruments, and two anatomic structures. Annotation of selected frames from robot-assisted minimally invasive esophagectomies was performed by at least three independent medical experts. To test whether AL reduces annotation effort, we performed a prospective annotation study comparing AL with equidistant sampling (EQS) for frame selection. Multiple Bayesian ResNet18 architectures were trained on a multicentric dataset, consisting of 22 videos from two centers. RESULTS: In total, 14,004 frames were tag annotated. A mean F1-score of 0.75 ± 0.16 was achieved for all features. The highest F1-score was achieved for the instruments (mean 0.80 ± 0.17). This result is also reflected in the inter-rater-agreement (1-rater-kappa > 0.82). Compared to EQS, AL showed better recognition results for the instruments with a significant difference in the McNemar test comparing correctness of predictions. Moreover, in contrast to EQS, AL selected more frames of the four less common instruments (1512 vs. 607 frames) and achieved higher F1-scores for common instruments while requiring less training frames. CONCLUSION: We presented ten surgomic features relevant for bleeding events in esophageal surgery automatically extracted from surgical video using ML. AL showed the potential to reduce annotation effort while keeping ML performance high for selected features. The source code and the trained models are published open source.

Germ line variant GFI1-36N affects DNA repair and sensitizes AML cells to DNA damage and repair therapy.

ABSTRACT: Growth factor independence 1 (GFI1) is a DNA-binding transcription factor and a key regulator of hematopoiesis. GFI1-36N is a germ line variant, causing a change of serine (S) to asparagine (N) at position 36. We previously reported that the GFI1-36N allele has a prevalence of 10% to 15% among patients with acute myeloid leukemia (AML) and 5% to 7% among healthy Caucasians and promotes the development of this disease. Using a multiomics approach, we show here that GFI1-36N expression is associated with increased frequencies of chromosomal aberrations, mutational burden, and mutational signatures in both murine and human AML and impedes homologous recombination (HR)-directed DNA repair in leukemic cells. GFI1-36N exhibits impaired binding to N-Myc downstream-regulated gene 1 (Ndrg1) regulatory elements, causing decreased NDRG1 levels, which leads to a reduction of O6-methylguanine-DNA-methyltransferase (MGMT) expression levels, as illustrated by both transcriptome and proteome analyses. Targeting MGMT via temozolomide, a DNA alkylating drug, and HR via olaparib, a poly-ADP ribose polymerase 1 inhibitor, caused synthetic lethality in human and murine AML samples expressing GFI1-36N, whereas the effects were insignificant in nonmalignant GFI1-36S or GFI1-36N cells. In addition, mice that received transplantation with GFI1-36N leukemic cells treated with a combination of temozolomide and olaparib had significantly longer AML-free survival than mice that received transplantation with GFI1-36S leukemic cells. This suggests that reduced MGMT expression leaves GFI1-36N leukemic cells particularly vulnerable to DNA damage initiating chemotherapeutics. Our data provide critical insights into novel options to treat patients with AML carrying the GFI1-36N variant.

Dose-dependent expression of GFI1 alters metabolism in the haematopoietic progenitors and MLL::AF9-induced leukaemic cells.

Growth factor independence 1 (GFI1) is a transcriptional repressor protein that plays an essential role in the differentiation of myeloid and lymphoid progenitors. We and other groups have shown that GFI1 has a dose-dependent role in the initiation, progression, and prognosis of acute myeloid leukaemia (AML) patients by inducing epigenetic changes. We now demonstrate a novel role for dose-dependent GFI1 expression in regulating metabolism in haematopoietic progenitor and leukaemic cells. Using in-vitro and ex-vivo murine models of MLL::AF9-induced human AML and extra-cellular flux assays, we now demonstrate that a lower GFI1 expression enhances oxidative phosphorylation rate via upregulation of the FOXO1- MYC axis. Our findings underscore the significance of therapeutic exploitation in GFI1-low-expressing leukaemia cells by targeting oxidative phosphorylation and glutamine metabolism.

A clinical screening tool to detect genetic cancer predisposition in pediatric oncology shows high sensitivity but can miss a substantial percentage of affected children.

PURPOSE: Clinical checklists are the standard of care to determine whether a child with cancer shows indications for genetic testing. Nevertheless, the efficacy of these tests to reliably detect genetic cancer predisposition in children with cancer is still insufficiently investigated. METHODS: We assessed the validity of clinically recognizable signs to identify cancer predisposition by correlating a state-of-the-art clinical checklist to the corresponding exome sequencing analysis in an unselected single-center cohort of 139 child-parent data sets. RESULTS: In total, one-third of patients had a clinical indication for genetic testing according to current recommendations, and 10.1% (14 of 139) of children harbored a cancer predisposition. Of these, 71.4% (10 of 14) were identified through the clinical checklist. In addition, >2 clinical findings in the checklist increased the likelihood to identifying genetic predisposition from 12.5% to 50%. Furthermore, our data revealed a high rate of genetic predisposition (40%, 4 of 10) in myelodysplastic syndrome cases, while no (likely) pathogenic variants were identified in the sarcoma and lymphoma group. CONCLUSION: In summary, our data show high checklist sensitivity, particularly in identifying childhood cancer predisposition syndromes. Nevertheless, the checklist used here also missed 29% of children with a cancer predisposition, highlighting the drawbacks of sole clinical evaluation and underlining the need for routine germline sequencing in pediatric oncology.

A Carboxy-terminal Smarcb1 Point Mutation Induces Hydrocephalus Formation and Affects AP-1 and Neuronal Signalling Pathways in Mice.

The BAF (BRG1/BRM-associated factor) chromatin remodelling complex is essential for the regulation of DNA accessibility and gene expression during neuronal differentiation. Mutations of its core subunit SMARCB1 result in a broad spectrum of pathologies, including aggressive rhabdoid tumours or neurodevelopmental disorders. Other mouse models have addressed the influence of a homo- or heterozygous loss of Smarcb1, yet the impact of specific non-truncating mutations remains poorly understood. Here, we have established a new mouse model for the carboxy-terminal Smarcb1 c.1148del point mutation, which leads to the synthesis of elongated SMARCB1 proteins. We have investigated its impact on brain development in mice using magnetic resonance imaging, histology, and single-cell RNA sequencing. During adolescence, Smarcb11148del/1148del mice demonstrated rather slow weight gain and frequently developed hydrocephalus including enlarged lateral ventricles. In embryonic and neonatal stages, mutant brains did not differ anatomically and histologically from wild-type controls. Single-cell RNA sequencing of brains from newborn mutant mice revealed that a complete brain including all cell types of a physiologic mouse brain is formed despite the SMARCB1 mutation. However, neuronal signalling appeared disturbed in newborn mice, since genes of the AP-1 transcription factor family and neurite outgrowth-related transcripts were downregulated. These findings support the important role of SMARCB1 in neurodevelopment and extend the knowledge of different Smarcb1 mutations and their associated phenotypes.

Portal of Medical Data Models: Application in Federated Data Capture.

So far, the portal for medical data models allows its users to download medical forms in a standardized format. Importing data models into electronic data capture software involved a manual step of downloading and importing the files. Now, the portal was enhanced with a web services interface to allow electronic data capture systems to automatically download the forms. This mechanism can be used in federated studies to ensure that all partners are working with identical definitions of study forms.

Federated electronic data capture (fEDC): Architecture and prototype.

In clinical research as well as patient care, structured documentation of findings is an important task. In many cases, this is achieved by means of electronic case report forms (eCRF) using corresponding information technology systems. To avoid double data entry, eCRF systems can be integrated with electronic health records (EHR). However, when researchers from different institutions collaborate in collecting data, they often use a single joint eCRF system on the Internet. In this case, integration with EHR systems is not possible in most cases due to information security and data protection restrictions. To overcome this shortcoming, we propose a novel architecture for a federated electronic data capture system (fEDC). Four key requirements were identified for fEDC: Definitions of forms have to be available in a reliable and controlled fashion, integration with electronic health record systems must be possible, patient data should be under full local control until they are explicitly transferred for joint analysis, and the system must support data sharing principles accepted by the scientific community for both data model and data captured. With our approach, sites participating in a joint study can run their own instance of an fEDC system that complies with local standards (such as being behind a network firewall) while also being able to benefit from using identical form definitions by sharing metadata in the Operational Data Model (ODM) format published by the Clinical Data Interchange Standards Consortium (CDISC) throughout the collaboration. The fEDC architecture was validated with a working open-source prototype at five German university hospitals. The fEDC architecture provides a novel approach with the potential to significantly improve collaborative data capture: Efforts for data entry are reduced and at the same time, data quality is increased since barriers for integrating with local electronic health record systems are lowered. Further, metadata are shared and patient privacy is ensured at a high level.

Noncancer-related Secondary Findings in a Cohort of 231 Children With Cancer and Their Parents.

Application of next-generation sequencing may lead to the detection of secondary findings (SF) not related to the initially analyzed disease but to other severe medically actionable diseases. However, the analysis of SFs is not yet routinely performed. We mined whole-exome sequencing data of 231 pediatric cancer patients and their parents who had been treated in our center for the presence of SFs. By this approach, we identified in 6 children (2.6%) pathogenic germline variants in 5 of the noncancer-related genes on the American College of Medical Genetics and Genomics (ACMG) SF v3.0 list, of which the majority were related to cardiovascular diseases ( RYR2 , MYBPC3 , KCNQ1 ). Interestingly, only the patient harboring the KCNQ1 variant showed at the time point of the analysis signs of the related Long QT syndrome. Moreover, we report 3 variants of unknown significance which, although not classified as pathogenic, have been reported in the literature to occur in individuals with the respective disease. While the frequency of patients with SFs is low, the impact of such findings on the patients' life is enormous, with regard to the potential prevention of life-threatening diseases. Hence, we are convinced that such actionable SF should be routinely analyzed.

Raising the standards of patient-centered outcomes research in myelodysplastic syndromes: Clinical utility and validation of the subscales of the QUALMS from the MDS-RIGHT project.

BACKGROUND: Clinical decision-making for patients with myelodysplastic syndromes (MDS) is challenging, and both disease and treatment effects heavily impact health-related quality of life (HRQoL) of these patients. Therefore, disease-specific HRQoL measures can be critical to harness the patient voice in MDS research. METHODS: We report a prospective international validation study of the Quality of Life in Myelodysplasia Scale (QUALMS) with a main focus on providing information on the psychometric characteristics of its three subscales: physical burden (QUALMS-P), emotional burden (QUALMS-E), and benefit finding (QUALMS-BF). The analysis is based on patients enrolled from three European countries and Israel, participating to the MDS-RIGHT Project. The scale structure and psychometric properties of the QUALMS were assessed. RESULTS: Overall, 270 patients with a median age of 74 years were analyzed and the majority of them (60.3%) had a low MDS-Comorbidity Index score. Results of the confirmatory factor analysis supported the underlying scale structure of the QUALMS, which, in addition to a total score, includes three subscales: QUALMS-P, QUALMS-E, and the QUALMS-BF. The QUALMS-P exhibited the highest Cronbach's alpha coefficients. Discriminant validity analysis indicated good results with the QUALMS-P and QUALMS-E distinguishing between patients with different performance status, comorbidity, anemia, and transfusion dependency status. No floor and ceiling effects were observed. Responsiveness to change analysis supported the validity of the measure. Patients with a hemoglobin (Hb) level of <11 g/dL at study entry, who subsequently showed an improvement in their Hb levels, also reported a mean score change of 9 and 8 points (scales ranging between 0 and 100) in the expected direction of the QUALMS-E and QUALMS-P, respectively. CONCLUSIONS: Our study provides additional validation data on the QUALMS from the international MDS-RIGHT Project. The use of this disease-specific HRQoL measure may contribute to raise quality standards of patient-centered outcomes research in MDS.

Transcriptome analyses in infertile men reveal germ cell-specific expression and splicing patterns.

The process of spermatogenesis-when germ cells differentiate into sperm-is tightly regulated, and misregulation in gene expression is likely to be involved in the physiopathology of male infertility. The testis is one of the most transcriptionally rich tissues; nevertheless, the specific gene expression changes occurring during spermatogenesis are not fully understood. To better understand gene expression during spermatogenesis, we generated germ cell-specific whole transcriptome profiles by systematically comparing testicular transcriptomes from tissues in which spermatogenesis is arrested at successive steps of germ cell differentiation. In these comparisons, we found thousands of differentially expressed genes between successive germ cell types of infertility patients. We demonstrate our analyses' potential to identify novel highly germ cell-specific markers (TSPY4 and LUZP4 for spermatogonia; HMGB4 for round spermatids) and identified putatively misregulated genes in male infertility (RWDD2A, CCDC183, CNNM1, SERF1B). Apart from these, we found thousands of genes showing germ cell-specific isoforms (including SOX15, SPATA4, SYCP3, MKI67). Our approach and dataset can help elucidate genetic and transcriptional causes for male infertility.

The CHK2 kinase is recurrently mutated and functionally impaired in the germline of pediatric cancer patients.

Predisposing CHEK2 germline variants are associated with various adult-type malignancies, whereas their impact on cancer susceptibility in childhood cancer is unclear. To understand the frequency of germline variants in the CHEK2 gene and their impact on pediatric malignancies, we used whole-exome sequencing to search for CHEK2 variants in the germlines of 418 children diagnosed with cancer in our clinics. Moreover, we performed functional analysis of the pathogenic CHEK2 variants to analyze the effect of the alterations on CHK2 protein function. We detected a CHEK2 germline variant in 32/418 (7.7%) pediatric cancer patients and 46.8% of them had leukemia. Functional analysis of the pathogenic variants revealed that 5 pathogenic variants impaired CHK2 protein function. 6/32 patients carried one of these clearly damaging CHEK2 variants and two of them harbored a matching family history of cancer. In conclusion, we detected germline CHEK2 variants in 7.7% of all pediatric cancer patients, of which a minority but still relevant fraction of approximately 20% had a profound impact on protein expression or its phosphorylation after irradiation-induced DNA damage. Accordingly, we conclude that CHEK2 variants increase the risk for not only adult-onset but also pediatric cancer.

Transcriptomic, Epigenomic, and Neuroanatomic Signatures Differ in Chronic Prurigo, Atopic Dermatitis, and Brachioradial Pruritus.

Scratching and scratch-induced injuries, including neuroanatomical alterations, are key characteristics of chronic pruritus entities of different origins. The aim of this study was to link gene expression (array hybridization, qPCR) with DNA methylation (array hybridization) and neuroanatomy (PGP9.5 staining) in chronic nodular prurigo (CNPG), atopic dermatitis (AD), brachioradial pruritus (BRP), and matched healthy controls. Specific signatures of gene expression and DNA methylation clearly discriminated pruritic lesional skin from nonpruritic skin in CNPG and from healthy skin of volunteers, respectively. Although intraepidermal nerve fiber density was indiscriminately reduced, the level of epidermal branching, assessed by a semiquantitative pattern analysis, differentiated the entities (CNPG > BRP > AD). Correspondingly, repellent SEMA3A showed the highest expression in AD, whereas axonal growth-promoting nerve GF was most prominent in CNPG and BRP. Overexpression of genes for nerve fiber regeneration (NELL2/NFKB/ARTN) was found in AD and CNPG but not in BRP. Our findings suggest that differential branching patterns rather than mere innervation density separate chronic itch conditions and reflect disease-specific local expression profiles. In pruritic dermatoses (AD and CNPG), nerve injury and subsequent sprouting may primarily result from chronic scratching, whereas genuine neuropathy is expected to underlie BRP.

Development and validation of an interpretable 3 day intensive care unit readmission prediction model using explainable boosting machines.

Background: Intensive care unit (ICU) readmissions are associated with mortality and poor outcomes. To improve discharge decisions, machine learning (ML) could help to identify patients at risk of ICU readmission. However, as many models are black boxes, dangerous properties may remain unnoticed. Widely used post hoc explanation methods also have inherent limitations. Few studies are evaluating inherently interpretable ML models for health care and involve clinicians in inspecting the trained model. Methods: An inherently interpretable model for the prediction of 3 day ICU readmission was developed. We used explainable boosting machines that learn modular risk functions and which have already been shown to be suitable for the health care domain. We created a retrospective cohort of 15,589 ICU stays and 169 variables collected between 2006 and 2019 from the University Hospital Münster. A team of physicians inspected the model, checked the plausibility of each risk function, and removed problematic ones. We collected qualitative feedback during this process and analyzed the reasons for removing risk functions. The performance of the final explainable boosting machine was compared with a validated clinical score and three commonly used ML models. External validation was performed on the widely used Medical Information Mart for Intensive Care version IV database. Results: The developed explainable boosting machine used 67 features and showed an area under the precision-recall curve of 0.119 ± 0.020 and an area under the receiver operating characteristic curve of 0.680 ± 0.025. It performed on par with state-of-the-art gradient boosting machines (0.123 ± 0.016, 0.665 ± 0.036) and outperformed the Simplified Acute Physiology Score II (0.084 ± 0.025, 0.607 ± 0.019), logistic regression (0.092 ± 0.026, 0.587 ± 0.016), and recurrent neural networks (0.095 ± 0.008, 0.594 ± 0.027). External validation confirmed that explainable boosting machines (0.221 ± 0.023, 0.760 ± 0.010) performed similarly to gradient boosting machines (0.232 ± 0.029, 0.772 ± 0.018). Evaluation of the model inspection showed that explainable boosting machines can be useful to detect and remove problematic risk functions. Conclusions: We developed an inherently interpretable ML model for 3 day ICU readmission prediction that reached the state-of-the-art performance of black box models. Our results suggest that for low- to medium-dimensional datasets that are common in health care, it is feasible to develop ML models that allow a high level of human control without sacrificing performance.