Renal Research Institute

Patient-reported outcome measures (PROMs) have become essential in contemporary internal medicine, where chronic, multisystem diseases and comorbidities make traditional biomedical endpoints insufficient as sole indicators of therapeutic benefit. PROMs capture symptom burden, functional capacity, emotional well-being, social participation and overall quality of life, thereby complementing laboratory indices, imaging findings and clinician-rated scales. Their growing use in regulatory evaluation, health technology assessment and value-based care underscores the need for rigorous methodology in their development, validation and interpretation. This review outlines key conceptual and practical issues that must be addressed for PROMs to provide valid, interpretable and clinically meaningful information. It emphasizes the central importance of a clearly articulated conceptual framework grounded in literature, clinical expertise and qualitative research with patients. It summarizes best practices in item generation and refinement, scaling and response options and the assessment of psychometric properties, including reliability, validity, responsiveness and interpretability. Particular attention is given to cross-cultural adaptation, differential item functioning and longitudinal measurement, which are crucial in heterogeneous internal medicine populations. The review also addresses the integration of PROMs into clinical trials and routine care, focusing on issues of feasibility, respondent burden, missing data and the translation of score changes into clinically actionable decisions. By clarifying these methodological foundations, it aims to support clinicians, investigators and policymakers in choosing, implementing and interpreting PROMs so that they genuinely reflect patients' experiences and priorities and thereby advance the practice of patient-centered internal medicine.

Background and objectivesIron plays a critical role beyond erythropoiesis, yet the prognostic significance of iron deficiency (ID) independent of anemia remains poorly defined in the peritoneal dialysis (PD) population. This study aimed to evaluate the association between iron status, specifically transferrin saturation (TSAT), and mortality in PD patients, independent of hemoglobin levels.Design, setting, participants, and measurementsWe conducted a retrospective cohort study of 11,013 adults who initiated PD at a large US dialysis network between December 2004 and January 2011. Patients had at least 180 days on PD and baseline data on TSAT, ferritin, hemoglobin, albumin, and white blood cell count. The primary outcome was all-cause mortality. Broadly adjusted associations between iron parameters and mortality were assessed using Cox proportional hazards models and restricted cubic splines, with adjustments for demographic, clinical, treatment-related, and laboratory variables including hemoglobin and ESA use.ResultsIron deficiency, defined as TSAT ≤20%, was present in 10% of patients at PD initiation. The cohort was 54% male and 70% Caucasian, with a mean age of 55 years; 39% had diabetes. While 91% received erythropoiesis-stimulating agents, only 34% received IV iron. After comprehensive adjustment, TSAT ≤20% remained independently associated with increased mortality (adjusted HR: 1.26; 95% CI: 1.12-1.42). Spline analyses showed a sharp rise in mortality risk at TSAT levels below 25%. Ferritin was inconsistently associated with mortality risk. During follow-up, 2704 deaths occurred (24.6% of the cohort) over a median 440-day follow-up.ConclusionsIron deficiency is common in incident PD patients and is associated with increased mortality risk, independent of anemia. These findings challenge current anemia-centric treatment paradigms and suggest that iron status, particularly TSAT, should be routinely assessed in PD patients regardless of hemoglobin levels. A prospective, randomized trial is warranted to evaluate whether proactive iron management improves outcomes in this population.

The impact of heat exposure on patients with end-stage kidney disease (ESKD) is of growing concern in the context of climate change. In this study, we investigated the association of heat exposure with hospitalization and mortality, and how the risk of these adverse health outcomes varied by climate region in the US. We obtained hospitalization and mortality data for patients with ESKD receiving in-center hemodialysis treatment between 2012 and 2018 at Fresenius Kidney Care facilities located within the contiguous US. We used the treatment facility location to assign heat exposure using maximum universal thermal climate index temperature data. We conducted a space-time-stratified case-crossover study using conditional Poisson regression with distributed lag nonlinear models to examine the effects of heat exposure at the 95th percentile of the region-specific temperature distribution for lags of three days. Stratified analyses were run to assess differences in associations across nine climate regions and three latitude bands. The cumulative lag 0-3 risk of hospitalization associated with heat exposure was highest in the West (rate ratio [RR]: 1.099; 95% confidence interval [CI]: 1.041, 1.160), whereas the highest risk of mortality was observed in the Northwest region (RR: 1.097; 95% CI: 1.007, 1.195). We observed significant increases in the risk of hospitalization at the low- and mid-latitude bands and a significant increase in the risk of mortality in the mid-latitude band. We observed spatial heterogeneity across US climate regions. The strongest effects of heat exposure were observed in the Ohio Valley, South, and West regions for hospitalization and the Upper Midwest, Southeast, and Northwest regions for mortality. Findings may be used to inform targeted interventions to patients with ESKD residing in areas with higher risks of adverse health outcomes following heat exposure.

This review explores the remarkable metabolic adaptations of species that thrive in extreme environments, providing insights into their resilience, flexibility and disease resistance. Species such as hibernating brown bears, migratory birds, cavefish, Greenland sharks and naked mole rats exhibit unique metabolic traits that challenge conventional paradigms of metabolic regulation. These adaptations, including resistance to hypoxia and metabolic ageing, offer potential solutions to human metabolic disorders, including obesity, type 2 diabetes and CVD. Insights from comparative physiology, particularly the mechanisms by which animals cope with food scarcity, extreme temperatures and hypoxia, could help identify novel therapeutic targets for advancing human health. For example, hibernation can serve as a model for understanding metabolic diseases, providing insights into reversible insulin resistance and energy homeostasis. This review also highlights the impact of environmental stressors, including climate change, on these species, which may jeopardise their survival despite their resilience. Accelerating anthropogenic environmental change threatens even the most resilient animal species. We call for a holistic approach to conservation and environmental protection to preserve these species and the valuable lessons they offer for managing our metabolic health.

Abstract Background and Aims Hemodialysis (HD) generates a vast amount of patient and machine data. These dynamic HD treatment data contain valuable information that can be used e.g., for the near-real-time prediction of intradialytic hypotension (IDH) (Zhang et al. NDT, 2023). To further accelerate data analysis, we explored the use of an autoencoder model, an unsupervised AI deep learning technique primarily used for extracting salient data features and to reduce data dimensionality. This technique is particularly effective (a) because it learns a representation that highlights the underlying patterns in the data while ignoring noise or irrelevant details, and (b) it compresses input data into a lower-dimensional latent space via an encoder and reconstructs it using a decoder, learning essential patterns by minimizing reconstruction loss. The lower-dimensional latent vector then could be used as input to other models, e.g., an IDH prediction model. Method We analyzed time series of intradialytic relative blood volume (RBV), arterial oxygen saturation (SaO2), and ultrafiltration volume (UFV). These data are recorded every 10 seconds by the HD machine and the integrated Crit-Line monitor (FME, Waltham, MA). We included treatments lasting at least 3 hours and comprising over 900 SaO2 recordings. The final dataset included 45,804 HD treatments from 22,218 patients, spanning Jan 30 to Feb 5, 2023. The dataset was divided into training (27,482 treatments, 60%), validation (9,161 treatments, 20%), and testing (9,161 treatments, 20%) subsets. The trajectories of RBV, SaO2, and UFV were normalized to a range of −1 to +1 using min-max normalization. An autoencoder (Fig 1), was developed. The encoder compresses intradialytic measurements of the three variables into a latent vector represented by a list of 25 numbers. The decoder then reconstructs data from the latent vector back to its original form. During the training process, the model minimizes the reconstruction loss, defined by mean squared error (MSE). Model performance is evaluated using mean absolute error (MAE) and normalized mean absolute error (NMAE) between the reconstructed and original data. Results Each of the three data channels (i.e., RBV, SaO2, UFV) contained 1,513 time steps. Per HD session, the encoder compressed the 3 × 1,513 = 4,539 elements into 25 elements. The three trajectories were then reconstructed from the latent vector through the decoder. The results (Fig 2) demonstrated a high similarity between the original trajectory (Ground Truth, blue) and the reconstructed trajectory (orange), indicating that the latent vector effectively preserves information of the original data and that the decoder successfully learns to reconstruct the trajectory, particularly for RBV and UFV. MAE and NMAE statistics are shown in Table 1. Conclusion The encoder-decoder model effectively compresses and reconstructs clinically relevant high-frequency hemodialysis time series trajectories, showcasing its potential as a robust tool for feature extraction. However, this method is unable to faithfully represent the high-frequency SaO2 oscillations. If and to what extent the use of the extracted features in models e.g., to predict IDH enables more accurate and efficient predictions warrants further exploration.

Abstract Background and Aims Aminoglycosides, such as Amikacin, are the cornerstone to treat severe infections in pediatric patients. Amikacin dosing requires achieving high peak concentrations to ensure efficacy against bacterial strains (up to > 60 mg/L depending on the strain) while maintaining trough levels < 2.5 mg/L to minimize nephrotoxic risks [1]. The minimum effective dose and frequency has been a debate due to the challenges of predicting ontogeny physiological changes. Current dosing schemes often miss peak and trough target levels in pediatric populations [1]. We present an age-stratified in-silico study to evaluate Amikacin dosing strategies in pediatric patients aged 2 to 24 months, focusing on optimizing peak and trough levels within a 24-hour dosing interval. Method An in-silico pharmacokinetic (PK) model based on [2] was validated against the clinical study results in [1]. Age-stratified virtual pediatric populations (2–24 months) were generated to account for maturational changes in GFR, weight, and height age-dependence based on growth charts and GFR data derived from (51)Cr-EDTA clearance measurements using a single blood sample method [3, 4]. Virtual populations were created for 2-, 6-, 12-, and 24-month-old cohorts (Table 1). The body surface area was computed using Mosteller's formula. Simulations in Python were conducted to evaluate Amikacin dosing strategies with 24-hour dosing intervals over 5 days of treatment. Results Using the standard maximal dosing scheme of 30 mg/kg once daily [1], the simulation shows that the fraction of patients with trough levels > 2.5 mg/L after 24 hours was 61.3%, 19.7%, 4.7% and 1.1% for the 2-, 6-, 12-, and 24-month age cohorts, respectively. Simulations using an optimized 24-hours dosing scheme, designed to maximize peak levels while maintaining trough levels below 2.5 mg/L, revealed a marked age-dependent difference in peak concentrations (Fig. 1). Median peak concentrations after initial administration were 42.5 mg/L, 48.9 mg/L, 50.9 mg/L, and 52.0 mg/L for the 2-, 6-, 12-, and 24-month age cohorts, respectively. A fraction of 6.2% of 2-month-olds and 18.8% of 24-month-olds reached peak levels >60 mg/L after the initial administration. Peak concentrations declined progressively across all age groups over the treatment days. The median peak concentration decreased by 56.4% for the 2-month-olds and 13.8% for the 24-month-olds by day 5 of treatment. Moreover, by day 5, none of the 2-month-olds and 11.4% of the 24-month-olds reached peak concentrations >60 mg/L. Conclusion Further dosage stratification of Amikacin by age is necessary in infants due to physiological kidney maturation and other developmental changes. High inter-patient variability emphasizes the need for therapeutic drug monitoring and dose adjustments during the treatment to optimize efficacy while minimizing the risk of nephrotoxicity.

Abstract Background and Aims Intradialytic hypotension (IDH) is a common complication of hemodialysis that negatively affects patients’ quality of life and increases mortality risk. One strategy to mitigate IDH involves lowering the ultrafiltration rate (UFR) to reduce dialysis strain; however, this often requires longer treatment times and may not benefit all patients. We developed an artificial intelligence (AI) model to predict IDH risk [1] and performed an in-silico simulation to characterize which patients are sensitive versus insensitive to changes in UFR. Our aim was to identify patient subgroups that could benefit from tailored UFR adjustments to reduce IDH events. Method We included 4,966 adult patients who underwent 1,187,298 hemodialysis sessions between 2021 and 2023 in Fresenius Medical Care Clinics in Czech Republic, Portugal, Spain, and Singapore (machine data available for these countries). Mean age was 68.9 ± 13.9 years, dialysis vintage 5.58 ± 5.73 years, and median UFR 7.0 [5.1–9.0] ml/hr/kg. The AI model used demographic, biochemical, and clinical treatment data. The final model demonstrated good discrimination (AUC = 0.75). Using model-based predictions, we then simulated the dose–response relationship between UFR and IDH risk per patient. We defined patients as “UFR-insensitive” when a substantial decrease in UFR (e.g., from the 85th to the 15th percentile) produced a small reduction in IDH risk (threshold 2.2%, i.e. IDH risk reduction median value) in more than 75% of their treatments. On the contrary, we defined patients as “UFR-sensitive” when a substantial decrease in UFR produced a substantial reduction in IDH risk (threshold 2.2%) in more than 75% of their treatments. We compared characteristics and observed IDH rates between UFR-insensitive and UFR-sensitive patients. Results Of the 4,966 patients analyzed, 1,177 were classified as UFR-insensitive and 1,394 as UFR-sensitive. Despite similar level of hydration status as judged by body composition monitoring, UFR-insensitive patients had fewer hypotensive events. Notably, they showed a negative change in systolic pressure from pre- to post-dialysis (versus a positive change in UFR-sensitive patients) and had lower pre-dialysis systolic pressures. UFR-insensitive patients also tended to have higher blood flow and blood volume during treatment, higher UFRs, greater small solute clearance, lower recirculation rates, and were less likely to have diabetes. Figure 1 illustrates simulated IDH risk curves at varying UFR values for a representative UFR-sensitive and UFR-insensitive patient. Conclusion Identifying patients whose IDH risk responds favorably to UFR adjustments is essential for optimizing dialysis treatment. Our findings suggest that UFR-sensitive patients experience more hypotensive events overall, highlighting the potential benefit of precisely targeted UFR modifications to reduce IDH. This approach could help improve dialysis outcomes and patient quality of life by tailoring treatment to individual risk profiles.

Abstract Background and Aims Gastrointestinal bleeding (GIB) is the most common bleeding event in dialysis,1 yet known incidence rates are based on small studies and the influence of patient characteristics is unknown. We used a nationally representative sample of dialysis patients treated in the United States (US) to characterize the incidence rates of major GIB episodes requiring hospitalization overall and by age, sex, and history of GIB comorbidity. Method We used data from a kidney care network from Jan 2018 through Mar 2021. Analysis included data from adult dialysis patients (age ≥ 18 years) who were treated with dialysis for ≥ 30 days. We excluded data from patients who started dialysis on or after Jan 2021 to provide a 3-month minimum follow up for outcomes to occur. GIB hospitalization was identified from international classification of diseases (ICD) diagnosis codes recorded as the primary, secondary, or tertiary discharge reason for hospitalization. ICD clusters defining GIB hospitalization and lesion location were based on the US Healthcare Cost and Utilization Project.2 We calculated the incidence of GIB hospitalization per 1,000 person-years (per 1000 py). Time at risk excluded GIB hospital admission time. Results Analysis included 366,839 adults on dialysis (mean age of 62.6 years, 57.7% male, 55.8% white race). A total of 25,057 patients (6.8%) experienced GIB hospitalization, with 18,407 (73.5%) having a single event and 6,650 (26.5%) experiencing recurrent events. The standardized incidence rate of GIB hospitalization was 52.6 per 1000 py (95% CI: 52.1–53.1). Among all GIB episodes, 6,623 were classified as upper GIB, 4,378 as lower GIB, with 2,905 episodes having an upper and lower GIB diagnosed during the same event. A remarkable number of episodes (n = 16,961) did not have a specified lesion location in the ICD code; there were no differences in patient characteristics for those with a known versus unknown lesion location. Assuming missingness at random, the standardized incidence rate of GIB hospitalization was higher for upper GIB (11.4 per 1000 py, 95% CI: 11.2–11.7) versus lower GIB events (7.5 per 1000 py, 95% CI: 7.3–7.7). The incidence rate of GIB hospitalization progressively increased with age, being about three times higher in individuals aged 75 and older versus 18–44 (Table 1). Females had a slight, but statistically significant higher GIB hospitalization incidence rate compared to males. A baseline GIB comorbidity was associated with a four times higher incidence rate of GIB hospitalization. Conclusion Major GIB events requiring hospitalization affect about 53 out of every 1000 people on dialysis every year in the US. In the general population, GIB hospitalization affects about 1 of every 1000 people each year. This more than 50-fold higher rate in the standardized incidence of GIB hospitalization reveals a remarkable burden and unmet need in the dialysis population.3 Sixty percent of GIB hospitalizations involved upper GIB, and 40% involved lower. Over 25% of patients had multiple GIB hospitalizations. Rates rose with age and were four times higher with prior GIB. These findings highlight the burden of GIB within this vulnerable population, emphasizing the need for improved strategies for detection, such as data driven prediction algorithms, and treatment before the need for hospitalization. This study’s findings should be interpreted considering certain limitations, including reliance on hospitalization and comorbidity data collected as part of routine dialysis therapy.