Development and evaluation of regression tree models for predicting in-hospital mortality of a national registry of COVID-19 patients over six pandemic surges.

BACKGROUND: Objective prognostic information is essential for good clinical decision making. In case of unknown diseases, scarcity of evidence and limited tacit knowledge prevent obtaining this information. Prediction models can be useful, but need to be not only evaluated on how well they predict, but also how stable these models are under fast changing circumstances with respect to development of the disease and the corresponding clinical response. This study aims to provide interpretable and actionable insights, particularly for clinicians. We developed and evaluated two regression tree predictive models for in-hospital mortality of COVID-19 patient at admission and 24 hours (24 h) after admission, using a national registry. We performed a retrospective analysis of observational routinely collected data. METHODS: Two regression tree models were developed for admission and 24 h after admission. The complexity of the trees was managed via cross validation to prevent overfitting. The predictive ability of the model was assessed via bootstrapping using the Area under the Receiver-Operating-Characteristic curve, Brier score and calibration curves. The tree models were assessed on the stability of their probabilities and predictive ability, on the selected variables, and compared to a full-fledged logistic regression model that uses variable selection and variable transformations using splines. Participants included COVID-19 patients from all ICUs participating in the Dutch National Intensive Care Evaluation (NICE) registry, who were admitted at the ICU between February 27, 2020, and November 23, 2021. From the NICE registry, we included concerned demographic data, minimum and maximum values of physiological data in the first 24 h of ICU admission and diagnoses (reason for admission as well as comorbidities) for model development. The main outcome measure was in-hospital mortality. We additionally analysed the Length-of-Stay (LoS) per patient subgroup per survival status. RESULTS: A total of 13,369 confirmed COVID-19 patients from 70 ICUs were included (with mortality rate of 28%). The optimism-corrected AUROC of the admission tree (with seven paths) was 0.72 (95% CI: 0.71-0.74) and of the 24 h tree (with 11 paths) was 0.74 (0.74-0.77). Both regression trees yielded good calibration and variable selection for both trees was stable. Patient subgroups comprising the tree paths had comparable survival probabilities as the full-fledged logistic regression model, survival probabilities were stable over six COVID-19 surges, and subgroups were shown to have added predictive value over the individual patient variables. CONCLUSIONS: We developed and evaluated regression trees, which operate at par with a carefully crafted logistic regression model. The trees consist of homogenous subgroups of patients that are described by simple interpretable constraints on patient characteristics thereby facilitating shared decision-making.

Prognostic models of in-hospital mortality of intensive care patients using neural representation of unstructured text: A systematic review and critical appraisal.

OBJECTIVE: To review and critically appraise published and preprint reports of prognostic models of in-hospital mortality of patients in the intensive-care unit (ICU) based on neural representations (embeddings) of clinical notes. METHODS: PubMed and arXiv were searched up to August 1, 2022. At least two reviewers independently selected the studies that developed a prognostic model of in-hospital mortality of intensive-care patients using free-text represented as embeddings and extracted data using the CHARMS checklist. Risk of bias was assessed using PROBAST. Reporting on the model was assessed with the TRIPOD guideline. To assess the machine learning components that were used in the models, we present a new descriptive framework based on different techniques to represent text and provide predictions from text. The study protocol was registered in the PROSPERO database (CRD42022354602). RESULTS: Eighteen studies out of 2,825 were included. All studies used the publicly-available MIMIC dataset. Context-independent word embeddings are widely used. Model discrimination was provided by all studies (AUROC 0.75-0.96), but measures of calibration were scarce. Seven studies used both structural clinical variables and notes. Model discrimination improved when adding clinical notes to variables. None of the models was externally validated and often a simple train/test split was used for internal validation. Our critical appraisal demonstrated a high risk of bias in all studies and concerns regarding their applicability in clinical practice. CONCLUSION: All studies used a neural architecture for prediction and were based on one publicly available dataset. Clinical notes were reported to improve predictive performance when used in addition to only clinical variables. Most studies had methodological, reporting, and applicability issues. We recommend reporting both model discrimination and calibration, using additional data sources, and using more robust evaluation strategies, including prospective and external validation. Finally, sharing data and code is encouraged to improve study reproducibility.

Knowledge, health beliefs and attitudes towards dementia and dementia risk reduction among descendants of people with dementia: a qualitative study using focus group discussions.

BACKGROUND: Individuals with a parental family history of dementia have an increased risk of developing dementia because they share their genes as well as their psychosocial behaviour. Due to this increased risk and their experience with dementia, they may be particularly eager to receive information regarding dementia risk reduction (DRR). This study evaluated the knowledge, beliefs and attitudes towards dementia and DRR among descendants of people with dementia. METHOD: Using a semi-structured topic guide, three focus group discussions were conducted consisting of 12 female (80%) and 3 male (20%) descendants of people with dementia with a mean (± SD) age of 48.8 (± 12) years. Focus group discussions were audio recorded and transcribed. Each transcript was analysed thoroughly, and where appropriate, a code was generated and assigned by two researchers independently. Then, similar codes were grouped together and categorized into themes. RESULTS: The items in the topic guide could only be addressed after participants had been given the opportunity to share their experiences of having a parent with dementia. Participants were unaware or uncertain about the possibility of reducing the risk of developing dementia and therefore hesitant to assess their dementia risk without treatment options in sight. Moreover, participants indicated that their general practitioner only gave some information on heritability, not on DRR. Although participants identified a large number of modifiable risk factors as a group during the group discussions, they were eager to receive more information on dementia and DRR. In the end, participants adopted a more positive attitude towards a DRR programme and provided suggestions for the development of future DRR programmes. CONCLUSIONS: Although the research aim was to evaluate the knowledge, beliefs and attitudes towards dementia and DRR, sharing experiences of having a parent with dementia seemed a prerequisite for considering participants' own risk of developing dementia and participating in a DRR programme. Knowledge of dementia and DRR was limited. Due to unawareness of the possibility of reducing dementia risk, participants were hesitant about assessing their dementia risk. Group discussions positively changed the perception of dementia risk assessment and participants' willingness to participate in a DRR programme.

Risk factors for the recurrence of obstetrical anal sphincter injury and the role of a mediolateral episiotomy: an analysis of a national registry.

OBJECTIVE: The assessment of risk factors, including mediolateral episiotomy (MLE), for the recurrence of obstetric anal sphincter injury (rOASI). DESIGN: Population-based cohort study. SETTING: Data from the nationwide database of the Dutch Perinatal Registry (Perined). POPULATION: A cohort of 391 026 women at term, of whom 9943 had an OASI in their first delivery and had a second vaginal delivery of a liveborn infant in cephalic position. METHODS: Possible risk factors were tested for statistical significance using univariate and multivariate logistic regression analysis. MAIN OUTCOME MEASURES: Rate of rOASI. RESULTS: The rate of rOASI was 5.8%. Multivariate analysis identified a birthweight of ≥4000 g (adjusted OR, aOR, 2.1, 95% CI 1.6-2.6) and a duration of second stage of ≥30 minutes (aOR 1.8, 95% CI 1.4-2.3) as statistically significant risk factors for rOASI. Mediolateral episiotomy was associated with a statistically significant lower rate of rOASI in spontaneous vaginal delivery (SVD) (aOR 0.4, 95% CI 0.3-0.5) and in operative vaginal delivery (OVD) (aOR 0.2, 95% CI 0.1-0.5). CONCLUSIONS: Women with a history of OASI have a higher rate of OASI in their next delivery. Duration of the second stage of ≥30 minutes and a birthweight of ≥4000 g are significantly associated with an increased rate of rOASI. Mediolateral episiotomy is associated with a significantly lower rate of rOASI in both SVD and OVD. TWEETABLE ABSTRACT: Mediolateral episiotomy is associated with a significant lower recurrence rate of OASI in women with an OASI in their first delivery.

The risk of intrapartum/neonatal mortality and morbidity following birth at 37 weeks of gestation: a nationwide cohort study.

OBJECTIVE: To assess intrapartum/neonatal mortality and morbidity risk in infants born at 37 weeks of gestation compared with infants born at 39-41 weeks of gestation. DESIGN: Nationwide cohort study. SETTING: The Netherlands. POPULATION: A total of 755 198 women delivering at term of a singleton without congenital malformations during 2010-14. METHODS: We used data from the national perinatal registry (PERINED). Analysis was performed with logistic regression and stratification for the way labour started and type of care. MAIN OUTCOME MEASURES: Intrapartum or neonatal mortality up to 28 days and adverse neonatal outcome (neonatal mortality, 5-minute Apgar <7, and/or neonatal intensive care unit admission). RESULTS: At 37 weeks of gestation intrapartum/neonatal mortality was 1.10‰ compared with 0.59‰ at 39-41 weeks (P < 0.0001). Adjusted odds ratio (aOR) for 37 weeks compared with 39-41 weeks was 1.84 (95% CI) 1.39-2.44). Adverse neonatal outcome at 37 weeks was 21.4‰ compared with 12.04‰ at 39-41 weeks (P < 0.0001) with an aOR 1.63 (95% CI 1.53-1.74). Spontaneous start of labour at 37 weeks of gestation was significantly associated with increased intrapartum/neonatal mortality with an aOR of 2.20 (95% CI 1.56-3.10), in both primary (midwifery-led) care and specialist care. Neither induction of labour nor planned caesarean section showed increased intrapartum/neonatal mortality risk. CONCLUSIONS: Birth at 37 weeks of gestation is independently associated with a higher frequency of clinically relevant adverse perinatal outcomes than birth at 39-41 weeks. In particular, spontaneous start of labour at 37 weeks of gestation doubles the risk for intrapartum/neonatal mortality. Extra fetal monitoring is warranted. TWEETABLE ABSTRACT: Birth at 37 weeks of gestation gives markedly higher intrapartum/neonatal mortality risk than at 39-41 weeks, especially with spontaneous start of labour.

The effect of ICU-tailored drug-drug interaction alerts on medication prescribing and monitoring: protocol for a cluster randomized stepped-wedge trial.

BACKGROUND: Drug-drug interactions (DDIs) can cause patient harm. Between 46 and 90% of patients admitted to the Intensive Care Unit (ICU) are exposed to potential DDIs (pDDIs). This rate is twice as high as patients on general wards. Clinical decision support systems (CDSSs) have shown their potential to prevent pDDIs. However, the literature shows that there is considerable room for improvement of CDSSs, in particular by increasing the clinical relevance of the pDDI alerts they generate and thereby reducing alert fatigue. However, consensus on which pDDIs are clinically relevant in the ICU setting is lacking. The primary aim of this study is to evaluate the effect of alerts based on only clinically relevant interactions for the ICU setting on the prevention of pDDIs among Dutch ICUs. METHODS: To define the clinically relevant pDDIs, we will follow a rigorous two-step Delphi procedure in which a national expert panel will assess which pDDIs are perceived clinically relevant for the Dutch ICU setting. The intervention is the CDSS that generates alerts based on the clinically relevant pDDIs. The intervention will be evaluated in a stepped-wedge trial. A total of 12 Dutch adult ICUs using the same patient data management system, in which the CDSS will operate, were invited to participate in the trial. Of the 12 ICUs, 9 agreed to participate and will be enrolled in the trial. Our primary outcome measure is the incidence of clinically relevant pDDIs per 1000 medication administrations. DISCUSSION: This study will identify pDDIs relevant for the ICU setting. It will also enhance our understanding of the effectiveness of alerts confined to clinically relevant pDDIs. Both of these contributions can facilitate the successful implementation of CDSSs in the ICU and in other domains as well. TRIAL REGISTRATION: Nederlands Trial register Identifier: NL6762 . Registered November 26, 2018.

Adults with congenital heart disease: ready for mobile health?

PURPOSE: Mobile health (mHealth) could improve the outcome of grown-up patients with congenital heart disease (GUCH) and reduce their emergency care utilisation. Inappropriate use of mHealth, however, can lead to data overload for professionals and unnecessary data collection for patients, increasing the burden for both. We aimed to determine the clinical characteristics of patients with high emergency care utilisation and to test whether these patients were willing to start using mHealth. METHODS: Clinical characteristics and emergency care utilisation of consecutive GUCH patients who visited one of the two participating cardiologists at the outpatient clinic of the Academic Medical Centre in Amsterdam were studied retrospectively. All patients were approached to fill in an mHealth questionnaire. A frequency of three or more emergency visits in 5 years was defined as high emergency care utilisation. RESULTS: In total, 202 consecutive GUCH patients who visited one of the two participating cardiologists were studied. Median age was 41 years, 47% were male, and 51% were symptomatic. In the previous 5 years, 134 emergency visits were identified. Of all patients, 8% had high emergency care utilisation. High emergency care utilisation was associated with patients being symptomatic, using antiarrhythmic drugs or diuretics. In total, 75% of all patients with high emergency care utilisation were willing to start using mHealth. CONCLUSION: GUCH patients who are symptomatic, those on antiarrhythmic drug therapy and those on diuretics are suitable candidates for enrolment in future mHealth initiatives because of both high care utilisation and high motivation to start using mHealth.

Assessment of the quality of fall detection and management in primary care in the Netherlands based on the ACOVE quality indicators.

UNLABELLED: We determined adherence to nine fall-related ACOVE quality indicators to investigate the quality of management of falls in the elderly population by general practitioners in the Netherlands. Our findings demonstrate overall low adherence to these indicators, possibly indicating insufficiency in the quality of fall management. Most indicators showed a positive association between increased risk for functional decline and adherence, four of which with statistical significance. INTRODUCTION: This study aims to investigate the quality of detection and management of falls in the elderly population by general practitioners in the Netherlands, using the Assessing Care of Vulnerable Elders (ACOVE) quality indicators. METHODS: Community-dwelling persons aged 70 years or above, registered in participating general practices, were asked to fill in a questionnaire designed to determine general practitioner (GP) adherence to fall-related indicators. We used logistic regression to estimate the association between increased risk for functional decline-quantified by the Identification of Seniors At Risk for Primary Care score-and adherence. We then cross-validated the self-reported falls with medical records. RESULTS: Of the 950 elders responding to our questionnaire, only 10.6 % reported that their GP proactively asked them about falls. Of the 160 patients who reported two or more falls, or one fall for which they visited the GP, only 23.1 % had fall documentation in their records. Adherence ranged between 13.6 and 48.6 %. There was a significant positive association between the ISAR-PC scores and adherence in four QIs. Documentation of falls was highest (36.7 %) in patients whom the GP had proactively asked about falls. CONCLUSION: Based on patient self-reports, adherence to the ACOVE fall-related indicators was poor, suggesting that the quality of evaluation and management of falls in community-dwelling older persons in the Netherlands is poor. The documentation of falls and fall-related risk factors was also poor. However, for most QIs, adherence to them increased with the increase in the risk of functional decline.

Mobile health in adults with congenital heart disease: current use and future needs.

OBJECTIVE: Many adults with congenital heart disease (CHD) are affected lifelong by cardiac events, particularly arrhythmias and heart failure. Despite the care provided, the cardiac event rate remains high. Mobile health (mHealth) brings opportunities to enhance daily monitoring and hence timely response in an attempt to improve outcome. However, it is not known if adults with CHD are currently using mHealth and what type of mHealth they may need in the near future. METHODS: Consecutive adult patients with CHD who visited the outpatient clinic at the Academic Medical Center in Amsterdam were asked to fill out questionnaires. Exclusion criteria for this study were mental impairment or inability to read and write Dutch. RESULTS: All 118 patients participated (median age 40 (range 18-78) years, 40 % male, 49 % symptomatic) and 92 % owned a smartphone. Whereas only a small minority (14 %) of patients used mHealth, the large majority (75 %) were willing to start. Most patients wanted to use mHealth in order to receive more information on physical health, and advice on progression of symptoms or signs of deterioration. Analyses on age, gender and complexity of defect showed significantly less current smartphone usage at older age, but no difference in interest or preferences in type of mHealth application for the near future. CONCLUSION: The relatively young adult CHD population only rarely uses mHealth, but the majority are motivated to start using mHealth. New mHealth initiatives are required in these patients with a chronic condition who need lifelong surveillance in order to reveal if a reduction in morbidity and mortality and improvement in quality of life can be achieved.

Factors that drive insulin-dosing decisions of diabetes care providers: a vignette-based study in the Netherlands.

AIM: To test how certain patient factors would influence the decision of Dutch care providers regarding insulin dose adjustments. We hypothesize that some of these decisions would diverge from recent evidence and consensus statements. METHODS: We developed narrative vignettes describing clinical scenarios of patients receiving basal insulin therapy. For each vignette, the respondents were asked to indicate whether they would advise a change in insulin dose. A total of 520 paper questionnaires were distributed among physicians and nurses in primary and secondary care in the Netherlands. Multivariate linear and logistic regression analyses were performed to identify factors associated with dosing decisions. RESULTS: A total of 190 (37%) questionnaires were returned. In cases of a severe rather than mild hypoglycaemic event, care providers were nearly five times more likely to decrease the dose (odds ratio 4.77, 95% CI 1.65-13.75). Care providers were six times more likely to increase the dose when the patient's current dose was low (30 units) rather than high (90 units) (odds ratio 6.38, 95% CI 3.04-13.37). The plasma glucose concentration during a hypoglycaemic event and a known history of cardiovascular disease did not influence the care providers' dosing decisions. CONCLUSION: Evidence regarding the optimum insulin titration is not always translated into clinical practice. When formulating guidelines, misconceptions should be identified and addressed.

Transient versus persistent acute kidney injury and the diagnostic performance of fractional excretion of urea in critically ill patients.

AIMS: To evaluate the performance of fractional excretion of urea (FeU) for differentiating transient (T) from persistent (P) acute kidney injury (AKI) and to assess performance of FeU in predicting AKI in patients admitted to the ICU. METHODS: We performed secondary analysis of a multicenter prospective observational cohort study on the predictive performance of biological markers for AKI in critically ill patients. AKI was diagnosed according to RIFLE staging. RESULTS: Of 150 patients, 51 and 41 patients were classified as having T-AKI and P-AKI, respectively. The diagnostic performance for FeU to discriminate T-AKI from P-AKI on the day of AKI was poor (AUC-ROC = 0.61; 95% CI: 0.49-0.73). The diagnostic performance of FeU to predict AKI 1 and 2 days prior to AKI was poor as well (AUC-ROC = 0.61; 95% CI: 0.47-0.74, and 0.58; 95% CI: 0.43-0.73, respectively). CONCLUSIONS: FeU does not seem to be helpful in differentiating T- from P-AKI in critically ill patients and it is a poor predictor of AKI.

Risk indicators for obstetrical anal sphincter injury in vaginal birth after caesarean section compared to first vaginal delivery.

OBJECTIVES: Comparison of the rate of obstetric anal sphincter injury (OASI) between women having their first vaginal birth after caesarean section (CS) and true nulliparous women with a vaginal delivery. Assessment of risk indicators for OASI in women with vaginal birth after one CS (VBAC). STUDY DESIGN: 28 535 women with their first VBAC and a cohort of 275 439 nulliparous women with a vaginal delivery of a liveborn infant in a cephalic position from the Dutch perinatal registry were analyzed. We compared the OASI rate with univariate and multivariate analysis. In women with VBAC possible risk indicators for OASI were assessed using univariate and multivariate logistic regression analysis. RESULTS: The rate of OASI was 5.2% in women with vaginal birth after CS and 4.0% in women with a first vaginal delivery. The adjusted OR (aOR) for vaginal birth after an elective CS was higher (aOR 1.34, 95% CI 1.23-1.47) compared to vaginal birth after an emergency CS (aOR 1.16, 95% CI 1.08-1.25). In women with vaginal birth after emergency CS, the aOR for the indication non-progressive labor was 1.18 (95% CI 1.08-1.29), whereas CS for suspected fetal distress was not significantly associated with obstetric anal sphincter injury in VBAC. In the 28 535 women with a VBAC, mediolateral episiotomy (MLE), birth weight < 3000 g and maternal age < 25 years were associated with a significantly lower rate of OASI. A gestational age of 42 weeks, birth weight ≥ 3500 g, operative vaginal delivery and duration of the 2nd stage of labour of ≥ 60 min were associated with a significantly higher rate of OASI. CONCLUSIONS: Women with a VBAC have a higher rate of OASI in comparison with women with a first vaginal delivery, with the exception of women with a vaginal birth after an emergency CS for suspected fetal distress. Factors associated with a significantly lower rate for OASI were MLE, maternal age < 25 and birth weight < 3000 g. A gestational age of 42 weeks, birth weight between 3500 and 4000 g and ≥ 4000 g, operative vaginal delivery and duration of the 2nd stage of delivery longer dan 60 min were associated with a significantly higher rate of OASI.

Recurrence risk of preterm birth in subsequent twin pregnancy after preterm singleton delivery.

OBJECTIVE: To determine the risk of preterm birth in a subsequent twin pregnancy after previous singleton preterm birth. DESIGN: Cohort study. SETTING: Nationwide study in the Netherlands. POPULATION: In all, 4071 nulliparous women who had a singleton delivery followed by a subsequent twin delivery between the years 1999 and 2007 were studied. METHODS: Outcome of subsequent twin pregnancy of women with a history of preterm singleton delivery was compared with pregnancy outcome of women with a history of term singleton delivery. First deliveries were subdivided into iatrogenic and spontaneous preterm deliveries. Furthermore analyses were performed by subgroups for gestational age at the time of singleton delivery. MAIN OUTCOME MEASURE: Spontaneous preterm birth (<37 weeks of gestation) in subsequent twin pregnancy. RESULTS: In the index singleton pregnancy, preterm birth occurred in 232 (5.7%) of 4071 women. The risk of subsequent twin preterm birth was significantly higher after previous singleton preterm delivery (56.9 versus 20.9%; odds ratio 5.0; 95% CI 3.8-6.6). Risk of subsequent twin preterm birth was dependent on the severity of previous singleton preterm birth and was highest after preceding spontaneous instead of iatrogenic singleton preterm delivery. CONCLUSION: Preterm birth of a singleton gestation is associated with an increased risk of spontaneous preterm birth in a subsequent twin pregnancy.

Identifying prognostic factors for clinical outcomes and costs in four high-volume surgical treatments using routinely collected hospital data.

Identifying prognostic factors (PFs) is often costly and labor-intensive. Routinely collected hospital data provide opportunities to identify clinically relevant PFs and construct accurate prognostic models without additional data-collection costs. This multicenter (66 hospitals) study reports on associations various patient-level variables have with outcomes and costs. Outcomes were in-hospital mortality, intensive care unit (ICU) admission, length of stay, 30-day readmission, 30-day reintervention and in-hospital costs. Candidate PFs were age, sex, Elixhauser Comorbidity Score, prior hospitalizations, prior days spent in hospital, and socio-economic status. Included patients dealt with either colorectal carcinoma (CRC, n = 10,254), urinary bladder carcinoma (UBC, n = 17,385), acute percutaneous coronary intervention (aPCI, n = 25,818), or total knee arthroplasty (TKA, n = 39,214). Prior hospitalization significantly increased readmission risk in all treatments (OR between 2.15 and 25.50), whereas prior days spent in hospital decreased this risk (OR between 0.55 and 0.95). In CRC patients, women had lower risk of in-hospital mortality (OR 0.64), ICU admittance (OR 0.68) and 30-day reintervention (OR 0.70). Prior hospitalization was the strongest PF for higher costs across all treatments (31-64% costs increase/hospitalization). Prognostic model performance (c-statistic) ranged 0.67-0.92, with Brier scores below 0.08. R-squared ranged from 0.06-0.19 for LoS and 0.19-0.38 for costs. Identified PFs should be considered as building blocks for treatment-specific prognostic models and information for monitoring patients after surgery. Researchers and clinicians might benefit from gaining a better insight into the drivers behind (costs) prognosis.

Interaction between geriatric syndromes in predicting three months mortality risk.

OBJECTIVES: Capturing frailty using a quick tool has proven to be challenging. We hypothesise that this is due to the complex interactions between frailty domains. We aimed to identify these interactions and assess whether adding interactions between domains improves mortality predictability. METHODS: In this retrospective cohort study, we selected all patients aged 70 or older who were admitted to one Dutch hospital between April 2015 and April 2016. Patient characteristics, frailty screening (using VMS (Safety Management System), a screening tool used in Dutch hospital care), length of stay, and mortality within three months were retrospectively collected from electronic medical records. To identify predictive interactions between the frailty domains, we constructed a classification tree with mortality as the outcome using five variables: the four VMS-domains (delirium risk, fall risk, malnutrition, physical impairment) and their sum. To determine if any domain interactions were predictive for three-month mortality, we performed a multivariable logistic regression analysis. RESULTS: We included 4,478 patients. (median age: 79 years; maximum age: 101 years; 44.8% male) The highest risk for three-month mortality included patients that were physically impaired and malnourished (23% (95%-CI 19.0-27.4%)). Subgroups had comparable three-month mortality risks based on different domains: malnutrition without physical impairment (15.2% (96%-CI 12.4-18.6%)) and physical impairment and delirium risk without malnutrition (16.3% (95%-CI 13.7-19.2%)). DISCUSSION: We showed that taking interactions between domains into account improves the predictability of three-month mortality risk. Therefore, when screening for frailty, simply adding up domains with a cut-off score results in loss of valuable information.

Can we reliably automate clinical prognostic modelling? A retrospective cohort study for ICU triage prediction of in-hospital mortality of COVID-19 patients in the Netherlands.

BACKGROUND: Building Machine Learning (ML) models in healthcare may suffer from time-consuming and potentially biased pre-selection of predictors by hand that can result in limited or trivial selection of suitable models. We aimed to assess the predictive performance of automating the process of building ML models (AutoML) in-hospital mortality prediction modelling of triage COVID-19 patients at ICU admission versus expert-based predictor pre-selection followed by logistic regression. METHODS: We conducted an observational study of all COVID-19 patients admitted to Dutch ICUs between February and July 2020. We included 2,690 COVID-19 patients from 70 ICUs participating in the Dutch National Intensive Care Evaluation (NICE) registry. The main outcome measure was in-hospital mortality. We asessed model performance (at admission and after 24h, respectively) of AutoML compared to the more traditional approach of predictor pre-selection and logistic regression. FINDINGS: Predictive performance of the autoML models with variables available at admission shows fair discrimination (average AUROC = 0·75-0·76 (sdev = 0·03), PPV = 0·70-0·76 (sdev = 0·1) at cut-off = 0·3 (the observed mortality rate), and good calibration. This performance is on par with a logistic regression model with selection of patient variables by three experts (average AUROC = 0·78 (sdev = 0·03) and PPV = 0·79 (sdev = 0·2)). Extending the models with variables that are available at 24h after admission resulted in models with higher predictive performance (average AUROC = 0·77-0·79 (sdev = 0·03) and PPV = 0·79-0·80 (sdev = 0·10-0·17)). CONCLUSIONS: AutoML delivers prediction models with fair discriminatory performance, and good calibration and accuracy, which is as good as regression models with expert-based predictor pre-selection. In the context of the restricted availability of data in an ICU quality registry, extending the models with variables that are available at 24h after admission showed small (but significantly) performance increase.

Trends in preterm birth: singleton and multiple pregnancies in the Netherlands, 2000-2007.

OBJECTIVE: Several studies have reported increasing trends in preterm birth in developed countries, mainly attributable to an increase in medically indicated preterm births. Our aim was to describe trends in preterm birth among singleton and multiple pregnancies in the Netherlands. DESIGN: Prospective cohort study. SETTING: Nationwide study. POPULATION: We studied 1,451,246 pregnant women from 2000 to 2007. METHODS: We assessed trends in preterm birth. We subdivided preterm birth into spontaneous preterm birth after premature prelabour rupture of membranes (pPROM), medically indicated preterm birth and spontaneous preterm birth without pPROM. We performed analyses separately for singletons and multiples. MAIN OUTCOME MEASURES: The primary outcome was preterm birth, defined as birth before 37 weeks of gestation, with very preterm birth (<32 weeks of gestation) being a secondary outcome. RESULTS: The risk of preterm birth was 7.7% and the risk of very preterm birth was 1.3%. In singleton pregnancies, the preterm birth risk decreased significantly from 6.4% to 6.0% (P < 0.0001), mainly as a result of the decrease in spontaneous preterm birth without pPROM (3.6-3.1%, P < 0.0001). In multiple pregnancies, the preterm birth risk increased significantly (47.3-47.7%, P = 0.047), mainly as a result of medically indicated preterm birth, which increased from 15.0% to 17.9% (P < 0.0001). CONCLUSION: In the Netherlands, the preterm birth risk in singleton pregnancies decreased significantly over the years. The trend of increasing preterm birth risk reported in other countries was only observed in (medically indicated) preterm birth in multiple pregnancies.

Travel time from home to hospital and adverse perinatal outcomes in women at term in the Netherlands.

OBJECTIVE: To study the effect of travel time, at the start or during labour, from home to hospital on mortality and adverse outcomes in pregnant women at term in primary and secondary care. DESIGN: Population-based cohort study from 2000 up to and including 2006. SETTING: The Netherlands Perinatal Registry. POPULATION: A total of 751,926 singleton term hospital births. METHODS: We assessed the impact of travel time by car, calculated from the postal code of the woman's residence to the 99 maternity units, on neonatal outcome. Logistic regression modelling with adjustments for gestational age, maternal age, parity, ethnicity, socio-economic status, urbanisation, tertiary care centres and volume of the hospital was used. MAIN OUTCOME MEASURES: Mortality (intrapartum, and early and late neonatal mortality) and adverse neonatal outcomes (mortality, Apgar <4 and/or admission to a neonatal intensive care unit). RESULTS: The mortality was 1.5 per 1000 births, and adverse outcomes occurred in 6.0 per 1000 births. There was a positive relationship between longer travel time (≥20 minutes) and total mortality (OR 1.17, 95% CI 1.002-1.36), neonatal mortality within 24 hours (OR 1.51, 95% CI 1.13-2.02) and with adverse outcomes (OR 1.27, 95% CI 1.17-1.38). In addition to travel time, both delivery at 37 weeks of gestation (OR 2.23, 95% CI 1.81-2.73) or 41 weeks of gestation (OR 1.52, 95% CI 1.29-1.80) increased the risk of mortality. CONCLUSIONS: A travel time from home to hospital of 20 minutes or more by car is associated with an increased risk of mortality and adverse outcomes in women at term in the Netherlands. These findings should be considered in plans for the centralisation of obstetric care.

Explaining heterogeneity of individual treatment causal effects by subgroup discovery: An observational case study in antibiotics treatment of acute rhino-sinusitis.

OBJECTIVES: Individuals may respond differently to the same treatment, and there is a need to understand such heterogeneity of causal individual treatment effects. We propose and evaluate a modelling approach to better understand this heterogeneity from observational studies by identifying patient subgroups with a markedly deviating response to treatment. We illustrate this approach in a primary care case-study of antibiotic (AB) prescription on recovery from acute rhino-sinusitis (ARS). METHODS: Our approach consists of four stages and is applied to a large dataset in primary care dataset of 24,392 patients suspected of suffering from ARS. We first identify pre-treatment variables that either confound the relationship between treatment and outcome or are risk factors of the outcome. Second, based on the pre-treatment variables we create Synthetic Random Forest (SRF) models to compute the potential outcomes and subsequently the causal individual treatment effect (ITE) estimates. Third, we perform subgroup discovery using the ITE estimates as outcomes to identify positive and negative responders. Fourth, we evaluate the predictive performance of the identified subgroups for predicting the outcome in two ways: the likelihood ratio test, and whether the subgroups are selected via the Akaike Information Criterion (AIC) using backward stepwise variable selection. We validate the whole modelling strategy by means of 10-fold-cross-validation. RESULTS: Based on 20 pre-treatment variables, four subgroups (three for positive responders and one for negative responders) were identified. The log likelihood ratio tests showed that the subgroups were significant. Variable selection using the AIC kept two of the four subgroups, one for positive responders and one for negative responders. As for the validation of the whole modelling strategy, all reported measures (the number of pre-treatment variables associated with the outcome, number of subgroups, number of subgroups surviving variable selection and coverage) showed little variation. CONCLUSIONS: With the proposed approach, we identified subgroups of positive and negative responders to treatment that markedly deviate from the mean response. The subgroups showed additive predictive value of the outcome. The modelling approach strategy was shown to be robust on this dataset. Our approach was thus able to discover understandable subgroups from observational data that have predictive value and which may be considered by the clinical users to get insight into who responds positively or negatively to a proposed treatment.