When performance is not enough-A multidisciplinary view on clinical decision support.

Scientific publications about the application of machine learning models in healthcare often focus on improving performance metrics. However, beyond often short-lived improvements, many additional aspects need to be taken into consideration to make sustainable progress. What does it take to implement a clinical decision support system, what makes it usable for the domain experts, and what brings it eventually into practical usage? So far, there has been little research to answer these questions. This work presents a multidisciplinary view of machine learning in medical decision support systems and covers information technology, medical, as well as ethical aspects. The target audience is computer scientists, who plan to do research in a clinical context. The paper starts from a relatively straightforward risk prediction system in the subspecialty nephrology that was evaluated on historic patient data both intrinsically and based on a reader study with medical doctors. Although the results were quite promising, the focus of this article is not on the model itself or potential performance improvements. Instead, we want to let other researchers participate in the lessons we have learned and the insights we have gained when implementing and evaluating our system in a clinical setting within a highly interdisciplinary pilot project in the cooperation of computer scientists, medical doctors, ethicists, and legal experts.

A single centre in-depth analysis of death with a functioning kidney graft and reasons for overall graft failure.

BACKGROUND: High numbers of unknown classifications and inconsistent methodologies in previous studies make the interpretation of causes leading to graft loss difficult. In addition, data on a holistic view looking at both death with a functioning graft (DWFG) and death-censored graft failure (DCGF) are sparse. METHODS: In this single-centre study we included 1477 adult kidney transplants performed between 1997 and 2017, of which all 286 DWFGs until the end of observation were analysed and causes for death assigned. Additionally, the results were compared with the causes of 303 DCGFs of the same cohort to evaluate the impact of causes for overall graft loss. RESULTS: The most frequent causes for DWFG were cardiovascular disease (CVD) in 30.8%, malignancy in 28.3% and infections in 21%. Only 9.4% of reasons for DWFG were unknown. Sudden death occurred in 40% (35/88) of patients classified as DWFG due to CVD. Overall graft loss was related to the effect of immunosuppression in 36.2% [infection 20.9% (123/589), malignancy 15.3% (90/589)] and CVD in 22.4% (132/589). In 27.4% (161/589), graft failure was associated with underimmunosuppression (rejection). For infections (60 DWFG, 63 DCGF) and CVD (88 DWFG, 44 DCGF), a considerable overlap was observed between DWFG and DCGF. For patients >70 years of age at transplantation, medical events accounted for 78% of overall graft losses and only 6.5% were associated with rejection. CONCLUSIONS: DWFG and DCGF share more causes for graft loss than previously reported and sudden death plays an underestimated role in death with a functioning graft.

Evaluation of a clinical decision support system for detection of patients at risk after kidney transplantation.

Patient care after kidney transplantation requires integration of complex information to make informed decisions on risk constellations. Many machine learning models have been developed for detecting patient outcomes in the past years. However, performance metrics alone do not determine practical utility. We present a newly developed clinical decision support system (CDSS) for detection of patients at risk for rejection and death-censored graft failure. The CDSS is based on clinical routine data including 1,516 kidney transplant recipients and more than 100,000 data points. In a reader study we compare the performance of physicians at a nephrology department with and without the CDSS. Internal validation shows AUC-ROC scores of 0.83 for rejection, and 0.95 for graft failure. The reader study shows that predictions by physicians converge toward the CDSS. However, performance does not improve (AUC-ROC; 0.6413 vs. 0.6314 for rejection; 0.8072 vs. 0.7778 for graft failure). Finally, the study shows that the CDSS detects partially different patients at risk compared to physicians. This indicates that the combination of both, medical professionals and a CDSS might help detect more patients at risk for graft failure. However, the question of how to integrate such a system efficiently into clinical practice remains open.

"Nothing works without the doctor:" Physicians' perception of clinical decision-making and artificial intelligence.

Introduction: Artificial intelligence-driven decision support systems (AI-DSS) have the potential to help physicians analyze data and facilitate the search for a correct diagnosis or suitable intervention. The potential of such systems is often emphasized. However, implementation in clinical practice deserves continuous attention. This article aims to shed light on the needs and challenges arising from the use of AI-DSS from physicians' perspectives. Methods: The basis for this study is a qualitative content analysis of expert interviews with experienced nephrologists after testing an AI-DSS in a straightforward usage scenario. Results: The results provide insights on the basics of clinical decision-making, expected challenges when using AI-DSS as well as a reflection on the test run. Discussion: While we can confirm the somewhat expectable demand for better explainability and control, other insights highlight the need to uphold classical strengths of the medical profession when using AI-DSS as well as the importance of broadening the view of AI-related challenges to the clinical environment, especially during treatment. Our results stress the necessity for adjusting AI-DSS to shared decision-making. We conclude that explainability must be context-specific while fostering meaningful interaction with the systems available.

Comparison of SARS-CoV-2 Antibody Response 4 Weeks After Homologous vs Heterologous Third Vaccine Dose in Kidney Transplant Recipients: A Randomized Clinical Trial.

Importance: Fewer than 50% of kidney transplant recipients (KTRs) develop antibodies against the SARS-CoV-2 spike protein after 2 doses of an mRNA vaccine. Preliminary data suggest that a heterologous vaccination, combining mRNA and viral vector vaccines, may increase immunogenicity. Objective: To assess the effectiveness of a third dose of an mRNA vs a vector vaccine in KTRs who did not have antibodies against the SARS-CoV-2 spike protein after 2 doses of an mRNA vaccine. Design, Setting, and Participants: This was a single center, single-blinded, 1:1 randomized clinical trial of a third dose of vaccine against SARS-CoV-2, conducted from June 15 to August 16, 2021, in 201 KTRs who had not developed SARS-CoV-2 spike protein antibodies after 2 doses of an mRNA vaccine. Data analyses were performed from August 17 to August 31, 2021. Interventions: mRNA (BNT162b2 or mRNA-1273) or vector (Ad26COVS1) as a third dose of a SARS-CoV-2 vaccine. Main Outcomes and Measures: The primary study end point was seroconversion after 4 weeks (29-42 days) following the third vaccine dose. Secondary end points included neutralizing antibodies and T-cell response assessed by interferon-γ release assays (IGRA). In addition, the association of patient characteristics and vaccine response was assessed using logistic regression, and the reactogenicity of the vaccines was compared. Results: Among the study population of 197 kidney transplant recipients (mean [SD] age, 61.2 [12.4] years; 82 [42%] women), 39% developed SARS-CoV-2 antibodies after the third vaccine. There was no statistically significant difference between groups, with an antibody response rate of 35% and 42% for the mRNA and vector vaccines, respectively. Only 22% of seroconverted patients had neutralizing antibodies. Similarly, T-cell response assessed by IGRA was low with only 17 patients showing a positive response after the third vaccination. Receiving nontriple immunosuppression (odds ratio [OR], 3.59; 95% CI, 1.33-10.75), longer time after kidney transplant (OR, 1.44; 95% CI, 1.15-1.83, per doubling of years), and torque teno virus plasma levels (OR, 0.92; 95% CI, 0.88-0.96, per doubling of levels) were associated with vaccine response. The third dose of an mRNA vaccine was associated with a higher frequency of local pain at the injection site compared with the vector vaccine, while systemic symptoms were comparable between groups. Conclusions and Relevance: This randomized clinical trial found that 39% of KTRs without an immune response against SARS-CoV-2 after 2 doses of an mRNA vaccine developed antibodies against the SARS-CoV-2 spike protein 4 weeks after a third dose of an mRNA or a vector vaccine. The heterologous vaccination strategy with a vector-based vaccine was well tolerated and safe but not significantly better than the homologous mRNA-based strategy. Trial Registration: EudraCT Identifier: 2021-002927-39.

Clinicopathologic Features and Risk Factors of Proteinuria in Transplant Glomerulopathy.

Background: Transplant glomerulopathy (TG) is one of the main causes of post-transplant proteinuria (PU). The features and possible risk factors for proteinuria in TG patients are uncertain. Methods: We investigated all patients who had biopsy-proven TG from 2000 to 2018 in our center. The clinical and histological data were compared between two groups with or without PU (cut-off = 0.3 g/day). Spearman correlation analysis was used to evaluate the relationship between PU and pathological changes. The risk factors for PU in TG patients were determined by multivariable logistic regression analysis. Results: One hundred and twenty-five (75.76%) of all enrolled 165 TG patients had proteinuria ≥0.3 g/day at the time of biopsy. TG patients' PU level was significantly correlated with Banff lesion score cg (ρ = 0.247, P = 0.003), and mm (ρ = 0.257, P = 0.012). Systolic blood pressure ≥140 mmHg (OR 2.72, 95% CI 1.04-7.10, P = 0.041), diastolic blood pressure ≥90 mmHg (OR 4.84, 95% CI 1.39-16.82, P = 0.013), peak PRA ≥5% (OR 6.47, 95% CI 1.67-25.01, P = 0.007), positive C4d staining (OR 4.55, 95% CI 1.29-16.11, 0.019), tacrolimus-based regimen (OR 3.5, 95% CI 1.28-9.54, P = 0.014), and calcium channel blocker usage (OR 4.38, 95% CI 1.59-12.09, P = 0.004) were independent risk factors for PU. Conclusions: Proteinuria is common in TG patients. systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, peak PRA ≥5%, positive C4d staining, tacrolimus-based regimen, and calcium channel blocker usage are associated with proteinuria in TG patients.

Accuracy of a Novel SARS-CoV-2 Antigen-Detecting Rapid Diagnostic Test from Standardized Self-Collected Anterior Nasal Swabs.

Background Antigen-detecting rapid diagnostic tests (Ag-RDT) for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) offer new opportunities for the quick and laboratory-independent identification of infected individuals for control of the SARS-CoV-2 pandemic. Despite the potential benefits, nasopharyngeal sample collection is frequently perceived as uncomfortable by patients and requires trained healthcare personnel with protective equipment. Therefore, anterior nasal self-sampling is increasingly recognized as a valuable alternative. Methods We performed a prospective, single-center, point of care validation of an Ag-RDT using a polypropylene absorbent collector for standardized self-collected anterior nasal swabs. Real-time polymerase chain reaction (RT-PCR) from combined oropharyngeal/nasopharyngeal swabs served as a comparator. Primary endpoint was sensitivity of the standardized Ag-RDT in symptomatic patients with medium or high viral concentration (≥1 million RNA copies on RT-PCR for SARS-CoV-2). Results Between 12 February and 22 March 2021, 388 participants were enrolled. After exclusion of 9 patients for which no PCR result could be obtained, the novel Ag-RDT was evaluated based on 379 participants, of whom 273 were symptomatic and 106 asymptomatic. In 61 samples from symptomatic patients with medium or high viral load (≥1 million RNA copies), the sensitivity of the standardized Ag-RDT was 96.7% (59/61; 95% confidence interval (CI): 88.7-99.6%) for the primary endpoint. In total, 62 positive Ag-RDT results were detected out of 70 RT-PCR positive individuals, yielding an overall sensitivity of 88.6% (95% CI: 78.7-94.9%). Specificity was 99.7% (95% CI: 98.2-100%) in 309 RT-PCR negative individuals. Conclusions Here, we present a validation of a novel Ag-RDT with a standardized sampling process for anterior nasal self-collection, which meets World Health Organisation (WHO) criteria of ≥80% sensitivity and ≥97% specificity. Although less sensitive than RT-PCR, this assay could be beneficial due to its rapid results, ease of use, and suitability for standardized self-testing.

TBase - an Integrated Electronic Health Record and Research Database for Kidney Transplant Recipients.

TBase is an electronic health record (EHR) for kidney transplant recipients (KTR) combining automated data entry of key clinical data (e.g., laboratory values, medical reports, radiology and pathology data) via standardized interfaces with manual data entry during routine treatment (e.g., clinical notes, medication list, and transplantation data). By this means, a comprehensive database for KTR is created with benefits for routine clinical care and research. It enables both easy everyday clinical use and quick access for research questions with highest data quality. This is achieved by the concept of data validation in clinical routine in which clinical users and patients have to rely on correct data for treatment and medication plans and thereby validate and correct the clinical data in their daily practice. This EHR is tailored for the needs of transplant outpatient care and proved its clinical utility for more than 20 years at Charité - Universitätsmedizin Berlin. It facilitates efficient routine work with well-structured, comprehensive long-term data and allows their easy use for clinical research. To this point, its functionality covers automated transmission of routine data via standardized interfaces from different hospital information systems, availability of transplant-specific data, a medication list with an integrated check for drug-drug interactions, and semi-automated generation of medical reports among others. Key elements of the latest reengineering are a robust privacy-by-design concept, modularity, and hence portability into other clinical contexts as well as usability and platform independence enabled by HTML5 (Hypertext Markup Language) based responsive web design. This allows fast and easy scalability into other disease areas and other university hospitals. The comprehensive long-term datasets are the basis for the investigation of Machine Learning algorithms, and the modular structure allows to rapidly implement these into clinical care. Patient reported data and telemedicine services are integrated into TBase in order to meet future needs of the patients. These novel features aim to improve clinical care as well as to create new research options and therapeutic interventions.

Exploring the Complexity of Death-Censored Kidney Allograft Failure.

BACKGROUND: Few studies have thoroughly investigated the causes of kidney graft loss (GL), despite its importance. METHODS: A novel approach assigns each persistent and relevant decline in renal function over the lifetime of a renal allograft to a standardized category, hypothesizing that singular or multiple events finally lead to GL. An adjudication committee of three physicians retrospectively evaluated indication biopsies, laboratory testing, and medical history of all 303 GLs among all 1642 recipients of transplants between January 1, 1997 and December 31, 2017 at a large university hospital to assign primary and/or secondary causes of GL. RESULTS: In 51.2% of the patients, more than one cause contributed to GL. The most frequent primary or secondary causes leading to graft failure were intercurrent medical events in 36.3% of graft failures followed by T cell-mediated rejection (TCMR) in 34% and antibody-mediated rejection (ABMR) in 30.7%. In 77.9%, a primary cause could be attributed to GL, of which ABMR was most frequent (21.5%). Many causes for GL were identified, and predominant causes for GL varied over time. CONCLUSIONS: GL is often multifactorial and more complex than previously thought.

Digital Home-Monitoring of Patients after Kidney Transplantation: The MACCS Platform.

The MACCS (Medical Assistant for Chronic Care Service) platform enables secure sharing of key medical information between patients after kidney transplantation and physicians. Patients provide information such as vital signs, well-being, and medication intake via smartphone apps. The information is transferred directly into a database and electronic health record at the kidney transplant center, which is used for routine patient care and research. Physicians can send an updated medication plan and laboratory data directly to the patient app via this secure platform. Other features of the app are medical messages and video consultations. Consequently, the patient is better-informed, and self-management is facilitated. In addition, the transplant center and the patient's local nephrologist automatically exchange notes, medical reports, laboratory values, and medication data via the platform. A telemedicine team reviews all incoming data on a dashboard and takes action, if necessary. Tools to identify patients at risk for complications are under development. The platform exchanges data via a standardized secure interface (Health Level 7 (HL7), Fast Healthcare Interoperability Resources (FHIR)). The standardized data exchange based on HL7 FHIR guarantees interoperability with other eHealth solutions and allows rapid scalability to other chronic diseases. The underlying data protection concept is in concordance with the latest European General Data Protection Regulation. Enrollment started in February 2020, and 131 kidney transplant recipients are actively participating as of July 2020. Two large German health insurance companies are currently funding the telemedicine services of the project. The deployment for other chronic kidney diseases and solid organ transplant recipients is planned. In conclusion, the platform is designed to enable home monitoring and automatic data exchange, empower patients, reduce hospitalizations, and improve adherence, and outcomes after kidney transplantation.

What happens after graft loss? A large, long-term, single-center observation.

The number of patients returning to dialysis after graft failure increases. Surprisingly, little is known about the clinical and immunological outcomes of this cohort. We retrospectively analyzed 254 patients after kidney allograft loss between 1997 and 2017 and report clinical outcomes such as mortality, relisting, retransplantations, transplant nephrectomies, and immunization status. Of the 254 patients, 49% had died 5 years after graft loss, while 27% were relisted, 14% were on dialysis and not relisted, and only 11% were retransplanted 5 years after graft loss. In the complete observational period, 111/254 (43.7%) patients were relisted. Of these, 72.1% of patients were under 55 years of age at time of graft loss and only 13.5% of patients were ≥65 years. Age at graft loss was associated with relisting in a logistic regression analysis. In the complete observational period, 42 patients (16.5%) were retransplanted. Only 4 of those (9.5%) were ≥65 years at time of graft loss. Nephrectomy had no impact on survival, relisting, or development of dnDSA. Patients after allograft loss have a high overall mortality. Immunization contributes to long waiting times. Only a very limited number of patients are retransplanted especially when ≥65 years at time of graft loss.

The relationship between proteinuria and allograft survival in patients with transplant glomerulopathy: a retrospective single-center cohort study.

Proteinuria and transplant glomerulopathy (TG) are common in kidney transplantation. To date, there is limited knowledge regarding proteinuria in different types of TG and its relationship to allograft survival. A retrospective cohort analysis of TG patients from indication biopsies was performed to investigate the relationship of proteinuria, histology, and graft survival. One hundred and seven (57.5%) out of 186 TG patients lost their grafts with a median survival of 14 [95% confidence interval (CI) 10-22] months after diagnosis. Proteinuria ≥1 g/24 h at the time of biopsy was detected in 87 patients (46.8%) and the median of proteinuria was 0.89 (range 0.05-6.90) g/24 h. TG patients with proteinuria ≥1 g/24 h had worse 5-year graft survival (29.9% vs. 53.5%, P = 0.001) compared with proteinuria <1 g/24 h. Proteinuria was associated with graft loss in univariable Cox regression [hazard ratio (HR) 1.25, 95% CI, 1.11-1.41, P < 0.001], and in multivariable analysis (adjusted HR 1.26, 95% CI 1.11-1.42, P < 0.001) independent of other risk factors including creatinine at biopsy, positive C4d, history of rejection, and Banff lesion score mesangial matrix expansion. In this cohort of TG patients, proteinuria at indication biopsy is common and associated with a higher proportion of graft loss.

eHealth in transplantation.

eHealth ("electronic" Health) is a new field in medicine that has the potential to change medical care, increase efficiency, and reduce costs. In this review, we analyzed the current status of eHealth in transplantation by performing a PubMed search over the last 5 years with a focus on clinical studies for post-transplant care. We retrieved 463 manuscripts, of which 52 clinical reports and eight randomized controlled trials were identified. Most studies were on kidney (n = 19), followed by liver (n = 10), solid organ (n = 7), bone-marrow (n = 6), and lung transplantation (n = 6). Eleven articles included adolescents/children. Investigated eHealth features covered the whole spectrum with mobile applications for patients (n = 24) and video consultations (n = 18) being most frequent. Prominent topics for patient apps were self-management (n = 16), adherence (n = 14), symptom-reporting (11), remote monitoring of vital signs (n = 8), educational (n = 7), and drug reminder (n = 7). In this review, we discuss opportunities and strengths of such new eHealth solutions, the implications for successful implementation into the healthcare process, the human factor, data protection, and finally, the need for better evidence from prospective clinical trials in order to confirm the claims on better patient care, potential efficiency gains and cost savings.