Meteen naar de content

Published articles

kik

Just published!

When accurate prediction models yield harmful self-fulfilling prophecies

Prediction models are popular in medical research and practice. However, this study shows that – when one uses predictive performance to monitor them – these models can actually lead to patient harm due to self-fulfilling prophecies.

The results call for a reconsideration of current development of these models in medical settings.

DOI: 10.1016/j.patter.2025.101229

Identifying a cohort of hospitalized chronic kidney disease patients

This study identified a chronic kidney disease cohort using electronic health records based on six logical rules.

Pooling resulted in identifying 17 805 hospitalized patients. The proportions identified by each logical rule varied from 2.1% to 8.4%. Our findings highlight the need to formalize the clinical definition into a computer-interpretable one, to avoid inconsistencies across studies.

DOI: 10.1093/ckj/sfaf073

Early lung cancer detection using Dutch primary care data

This study created two AI models to help detect lung cancer early by analyzing health information.

One model was based on these notes alone and the other combined the notes with other health data. Both models worked well in spotting patients at high risk for lung cancer. The model that included both the notes and other data was slightly more accurate.

DOI: 10.3390/cancers17071151

Reusable health data

FAIR principles in child and adolescent mental health research

This study identified barriers to implementing the FAIR principles in mental health research.

A total of 45 barriers were identified, including those related to organizational policies and software. Recommendations include adding a FAIR data steward to the research team, providing accessible guides, and ensuring sustainable funding to overcome these barriers.

DOI: 10.2196/59113

Anatomical Therapeutic Chemical (ATC) -to- RxNorm mapping

This study evaluates the completeness of mappings from ATC classification classes to RxNorm ingredient concepts.

The studied vocabulary repositories were OHDSI Standardized Vocabularies (OSV) and Unified Medical Language System (UMLS) Metathesaurus. OSV provides more mappings and is recommended, though UMLS may perform better in concrete applications.

DOI: 10.1016/j.ijmedinf.2024.105777

Resolvability, parsability, and consistency of resources

This study evaluated the quality of 16 linked resources related to rare diseases, which rely on Uniform Resource Identifiers (URIs) to connect and enrich data.

The authors applied six quality metrics to assess the resources and identified various types of quality issues, including malfunctioning URIs and undefined ontological terms. These insights inform the creation of guidelines aimed at enhancing the quality and interoperability of linked biomedical resources.

DOI: 10.1186/s1336-023-00299-3

Assessing the FAIRness of databases

This study assesses the findability, accessibility, interoperability, and reusability (FAIR) of two intensive care databases.

Two researchers applied the seventeen FAIRsFAIR metrics which were defined as minimal requirements for databases to be FAIR. The total scores achieved were 15.5 and 12.0 out of a maximum of 25. The main areas for development were the implementation of globally unique identifiers and standardizing linked metadata.

DOI: 10.1016/j.ijmedinf.2023.105104

Correctly structured problem lists in the EHR

This randomized controlled trial investigated the impact of structured problem lists in electronic health records (EHRs) on clinical decision-making, compared to information partly documented in the free text.

Results show that structured problem lists lead to better and faster decision-making among healthcare providers. The findings suggest increased use of structured lists and the development of implementation policies.

DOI: 10.1016/j.ijmedinf.2023.105264

Methods in Medical Informatics

Benchmark of out-of-distribution detection in medical data

Machine learning models often struggle with data distributions that differ from their training dataset.

It is important to detect these “out-of-distribution” (OOD) because prediction on such data can be unreliable. This study aims to evaluate and compare different OOD detection methods in the context of medical tabular data.

DOI:10.1016/j,ijmedinf.2024.105762

Increasing trends in a low 5-min Apgar score

This study analyzed the Dutch Perinatal Registry (PERINED) data (2010-2019) on 1,583,188 infants born ≥35 weeks of gestation.

A significant increase in the proportion of infants with low 5-minute Apgar scores (<7 and <4) was found. Instrumental vaginal delivery and emergency caesarean sections were performed less frequently, however these intervention groups showed the highest relative increase in infants with low Apgar scores.

DOI: 10.1038/s41372-023-01786-2

An integrated approach to geographic validation

This study validates a mortality prediction model for Transcatheter Aortic Valve Implantation across 16 hospitals using multicenter data.

Among 11,599 patients with 3.7% early mortality, performance varied widely between hospitals, with miscalibration in seven hospitals. Notable case-mix differences and limited model adaptability were evident. A subgroup displayed significant prediction errors and poor validation center performance.

DOI: 10.1016/j.jclinepi.2023.02.021

Validation of 30-day mortality prediction models for TAVI

To better understand the predictive performance of models for mortality after Transcatheter Aortic Valve Implantation, researchers explored the effect of time on two types of models (parametric and non-parametric).

They used data from the Netherlands Heart Registration between 2013 – 2019 (11,291 patients). Results showed that both models performed better when refitted over time.

DOI: 10.1016/j.heliyon.2023.e17139

Development of a prediction model for in-hospital falls

This study developed and validated two prediction models for falls among older hospital patients (2016-2021).

It included a dataset of 21,286 admissions. The first model included missing value predictors (Model-With) and the second model did not (Model-Without). Both models showed good calibration and fair discrimination. The Model-With performing better. The models use routine data and can potentially reduce nurses’ workload.

DOI: 10.1016/j.jamda.2023.03.006

Prolonged use of closed-loop inspired oxygen support

This study compared the effect of automated oxygen control (A-FiO2) with manual titration (M-FiO2) on non-invasive respiratory support in preterm infants.

The randomized controlled trial showed that A-FiO2 was better at maintaining the oxygen saturation within target range over a 28-days period but with a potential trade-off of increased time below the target range initially.

DOI: 10.1136/archdischild-2023-325831

Human Factors Engineering

Sustainable healthcare: development of a model

Sustainability is playing an increasingly important role in healthcare organizations, but improving the environmental impact in healthcare remains a challenge.

The aim of this research is to develop a model that allows healthcare professionals to assess and improve the sustainability performance of their organization or department.

Click here for more information.

Dermatologists’ experience with teledermatology

This study aims to understand how Dutch dermatologists experienced teledermatology.

Results of the questionnaire showed that dermatologists had a positive experience with teledermatology. However, the insufficient quality and incompleteness of the clinical content (photos and anamneses information) of the teledermatology consultation needs improvement.

DOI: 10.2196/56723

Environmental impact factors for sustainable healthcare

This review identified 360 impact factors to measure environmental impact in healthcare from 46 articles.

These factors were categorized resulting in the Healthcare Environmental Impact Factor (HEIF) scheme. HEIF can – for example – be used to select measurable indicators for quality management and guide future research in standardizing factors for cross-organization comparisons.


DOI: 10.3390/ijerph20186747

Reduction of the carbon footprint of surgical instrument trays.

This study used Integer Linear Programming (ILP) and Life Cycle Assessment (LCA) to guide reduction of the carbon footprint of surgical instrument trays.

By optimizing tray composition , a 46% annual reduction in emissions could be realized for radical hysterectomy procedures.

DOI: 10.1016/j.cesys.2023.100154

Quality of care (IT systems)

Independent association of organisational factors and ICU efficacy

This study looked at two different methods to measure the effect of organizational factors on ICU performance in Brazil and Uruguay.

Both methods, causal random forest and linear regression modelling, found that the number of nurses per ten beds improved ICU efficiency. Causal random forest was able to spot areas where overall results might not seem significant. This could help hospital managers to improve ICU efficiency.

DOI: 10.1016/j.ijmedinf.2024.105568

One-year transplant-free survival after ICU admission

This study investigates the long-term outcomes of ICU patients with acute decompensated liver cirrhosis or acute-on-chronic liver failure (ACLF).

In total, 47% of 3035 patients survived hospitalization, and the overall probability of 1-year transplant-free survival after hospital discharge was 0.61. The ACLF severity at ICU admission did not independently influence post-discharge survival.

DOI: 10.1016/j.jhep.2024.03.004

Quality improvement of Dutch ICUs 2009-2021

This study examined the quality of ICU care in the Netherlands (2009-2021) using data from the NICE registry.

In total, 705,822 admissions from 55 ICUs were included in the analyses. Results showed improvements in several indicators over time, such as shorter ICU stays and reduced in-hospital mortality. However, the COVID-19 pandemic disrupted some of these trends.

DOI: 10.1016/j.jcrc.2023.154461

Predicting 30-day mortality in ICU patients with a stroke.

This study developed and validated two predictive models to estimate the 30-day mortality for patients with a stroke admitted to the ICUs: one model for patients with ischaemic (N=8 422) and one model for patients with haemorrhagic stroke (N=5 881).

The 30-day mortality was 27% in the first group and 41% in the second group. Both models showed high discrimination (AUC of 0.85) and good calibration. 

DOI: 10.1097/EJA.0000000000001920

Strain on Scarce Intensive Care Beds

This study analyzed the impact of ICU resource allocation during the COVID-19 pandemic on a non-COVID-19 cohort.

Compared to the prepandemic period cohort (164.737 patients), the pandemic cohort had lower number of non-COVID patients (27% lower), fewer medical patients (3% lower), fewer comorbidities (4% lower), more vasoactive medication (6% higher), and a slightly higher case-mix adjusted hospital mortality (odds 1.08).

DOI: 10.1097/CCM.0000000000006156

Characteristics and outcome of COVID-19 ICU patients

This study reports mortality rates among patients with COVID-19 during the first wave and compared these rates with the following waves.

Between May 2020 and January 2023,  the mortality risks in the initial upsurge of the third wave was similar to the first wave, but mortality rates decreased in later periods.

DOI: 10.1186/s13613-023-01238-2

Epidemiology of kidney disease

Adult outcomes of childhood kidney replacement therapy 2008-2019

This study focused on young adults who began kidney replacement therapy during childhood. Almost 3000 patients were included.

Results showed that having a kidney transplant at age 18 led to better outcomes than being on dialysis, but these patients still had a shorter life expectancy compared to the general population. Between 18-23y, about 20% of the kidney transplant patients lost their graft.

DOI:10.1093/ndt/gfae189

Differences in kidney disease in men and women

This review reports on differences between men and women with kidney disease.

Women show a higher prevalence of chronic kidney disease stages 3-5, and men a higher prevalence of albuminuria. Women are less aware of their disease, and receive less screening, nephrologist care, and face greater barriers to kidney transplantation access. Men experience faster renal decline, higher mortality, and increased cardiovascular risk.

DOI: 10.10338/s4181-023-00784-z

PharmacoInformatics

The use of AI to optimize medication alerts

This scoping review identified 10 studies which used AI to optimize medication alerts in hospitals, with only 30% reporting both statistical and clinical outcomes.

Alerts optimized using AI-based methods resulted in a decreased alert burden, increased identification of inappropriate or atypical prescriptions, and enabled prediction of user responses.  Two of the 10 studies implemented AI alerts in hospitals, and none underwent external validation.

DOI: 10.1093/jamia/ocae076

The effect of CDSS for high-risk drug combinations

Using a cluster randomised stepped-wedge trial, a clinical decision support system (CDSS) only providing alerts for potential drug-drug interactions (DDIs) considered as high risk was implemented and evaluated in nine Dutch intensive care units.

The study found a 12% decrease in the number of administered high-risk drug combinations, indicating that tailoring alerts to the ICU setting improves CDSS effectiveness and patient safety.

DOI: 10.1016/S0140-6736(23)02465-0

Clinical notes for adverse drug event detection

A high-quality corpus of Dutch ICU medical notes was created to help computers detect medication side effects, especially those impacting the kidneys.

Two medical experts analyzed anonymized notes from 102 adult ICU patients suspected of having drug-related kidney issues. They labeled medications, medical problems, and their relationships using standardized, double-checked guidelines. Over 16,000 labels were added in total, including 614 for drug-related side effects.

DOI: 10.1007/s10579-025-09832-5

Drug-related causes attributed to acute kidney injury

This study examined causes of drug-related acute kidney injury and their documentation. Among 8,124 ICU admissions in Amsterdam UMC (2015-2020), 542 experienced acute kidney injury and 102 a drug-related acute kidney injury. 

Clinical notes documented all drug-related cases (100%), while the allergy module (1%) and diagnosis codes (0%) were (almost) not utilized.

DOI: 10.1016/j.jcrc.2023.154292

Acute kidney injury associated with nephrotoxic drugs

Acute Kidney Injury (AKI) due to nephrotoxic drugs is common in ICU patients.

Using real ICU data (92,616 admissions), this study identified 14 drug groups linked to higher risk of AKI. The groups included aminoglycosides, opioids and sympathomimetics. The results enhances current ICU practices and guides further research on drug-related AKI.

DOI: 10.1093/ckj/sfad160