Ensuring comparisons of health-care providers are fair: development and validation of risk prediction models for critically ill patients
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Background National clinical audit has a key role in ensuring quality in health care. When comparing outcomes between providers, it is essential to take the differing case mix of patients into account to make fair comparisons. Accurate risk prediction models are therefore required. Objectives To improve risk prediction models to underpin quality improvement programmes for the critically ill (i.e. patients receiving general or specialist adult critical care or experiencing an in-hospital cardiac arrest). Design Risk modelling study nested within prospective data collection. Setting Adult (general/specialist) critical care units and acute hospitals in the UK. Participants Patients admitted to an adult critical care unit and patients experiencing an in-hospital cardiac arrest attended by the hospital-based resuscitation team. Interventions None. Main outcome measures Acute hospital mortality (adult critical care); return of spontaneous circulation (ROSC) greater than 20 minutes and survival to hospital discharge (in-hospital cardiac arrest). Data sources The Case Mix Programme (adult critical care) and National Cardiac Arrest Audit (in-hospital cardiac arrest). Results The current Intensive Care National Audit & Research Centre (ICNARC) model was externally validated using data for 29,626 admissions to critical care units in Scotland (2007–9) and outperformed the Acute Physiology And Chronic Health Evaluation (APACHE) II model in terms of discrimination (c-index 0.848 vs. 0.806) and accuracy (Brier score 0.140 vs. 0.157). A risk prediction model for cardiothoracic critical care was developed using data from 17,002 admissions to five units (2010–12) and validated using data from 10,238 admissions to six units (2013–14). The model included prior location/urgency, blood lactate concentration, Glasgow Coma Scale (GCS) score, age, pH, platelet count, dependency, mean arterial pressure, white blood cell (WBC) count, creatinine level, admission following cardiac surgery and interaction terms, and it had excellent discrimination (c-index 0.904) and accuracy (Brier score 0.055). A risk prediction model for admissions to all (general/specialist) adult critical care units was developed using data from 155,239 admissions to 232 units (2012) and validated using data from 90,017 admissions to 216 units (2013). The model included systolic blood pressure, temperature, heart rate, respiratory rate, partial pressure of oxygen in arterial blood/fraction of inspired oxygen, pH, partial pressure of carbon dioxide in arterial blood, blood lactate concentration, urine output, creatinine level, urea level, sodium level, WBC count, platelet count, GCS score, age, dependency, past medical history, cardiopulmonary resuscitation, prior location/urgency, reason for admission and interaction terms, and it outperformed the current ICNARC model for discrimination and accuracy overall (c-index 0.885 vs. 0.869; Brier score 0.108 vs. 0.115) and across unit types. Risk prediction models for in-hospital cardiac arrest were developed using data from 14,688 arrests in 122 hospitals (2011–12) and validated using data from 7791 arrests in 143 hospitals (2012–13). The models included age, sex (for ROSC > 20 minutes), prior length of stay in hospital, reason for attendance, location of arrest, presenting rhythm, and interactions between rhythm and location. Discrimination for hospital survival exceeded that for ROSC > 20 minutes (c-index 0.811 vs. 0.720). Limitations The risk prediction models developed were limited by the data available within the current national clinical audit data sets. Conclusions We have developed and validated risk prediction models for cardiothoracic and adult (general and specialist) critical care units and for in-hospital cardiac arrest. Future work Future development should include linkage with other routinely collected data to enhance available predictors and outcomes. Funding details The National Institute for Health Research Health Services and Delivery Research programme.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it