Comparison of the Parsonnet score and hospital-specific models using cardiac surgery patients from Montreal
Bibliographic record
Abstract
Rationale: Risk assessment tools are frequently used to help surgeons and patients decide whether the potential benefits of cardiac surgery outweigh the risks. One of the most popular risk estimation tools, the 2000 Parsonnet score, has shown varying levels of predictive ability in different regions throughout the world. This risk estimation tool is based on 37 variables. According to the literature, institutional quality, medical personnel and patient ethnicity also have predictive value when measuring the risk of in-hospital mortality yet these are not included as risk factors in the model. It is hypothesized that a regional score based on the 2000 Parsonnet score would have greater accuracy in its patient population. This is because the 2000 Parsonnet score was developed using data from only one region (New Jersey) and the literature has shown that it has varying levels of accuracy in different regions. Methods: Patient level data from two Montreal hospitals were used to create two hospital-specific models based on the 2000 Parsonnet score using multivariable logistic regression. There were 1,162 patients included from one hospital and 2,656 patients from the other hospital. Risk factor coefficients of the hospital-specific models were compared to the coefficients of the Parsonnet model using z-tests. Additionally, the predictive accuracies of the hospital-specific models were compared to the Parsonnet model using receiver operating characteristic (ROC) analyses. The number of preventable deaths at different risk thresholds was also calculated between models.Results: Between the Montreal-based institutions, 2 variables had different risk factor coefficients. One hospital-specific model had 4 different risk factor weights and the other hospital-specific model also had 4 different weights when compared to the Parsonnet model, (p≤0.05). Only one model was found to be marginally more accurate with an area under the ROC curve of 0.882 compared to 0.868 for the Parsonnet model (p=0.049). No difference was detected for the other model with an area under the ROC curve of 0.804 compared to 0.798 for the Parsonnet model (p>0.05). Additionally, in one hospital, for three out of four risk thresholds, the use of the hospital-specific model instead of the Parsonnet model would have prevented between 2 and 8 in-hospital mortalities. In the other hospital, for four out of four risk thresholds, the use of the hospital-specific model instead of the Parsonnet model would have prevented between 3 and 18 in-hospital mortalities. Conclusions: These results demonstrate that the Parsonnet model has approximately the same risk factor weights and predictive accuracy as the hospital-specific models. Also, the potential clinical benefit of prevented mortality was too small to justify using hospital-specific models over the Parsonnet model. Future research in other regions and in a larger number of hospitals is required to validate this conclusion.
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.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.020 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".