Multivariable risk prediction model for neurosurgery

Multivariable risk prediction models are more accurate than clinicians estimate and are now used for surgical treatment of multiple diseases.
but lacks a multivariable risk prediction model for neurosurgery.
Currently, studies have used data from 175,313 neurosurgery cases in the National Surgical Quality Improvement Program (NSQIP) to determine the predictive factors for a 30-day prognosis for neurosurgery, including chronic obstructive pulmonary disease, congestive heart failure, assats of the American Society of Anesthesiologists (ASA-PS), and wound division.
NSQIP risk prediction model was internally validated in neurosurgery patients in the United States and evaluated by a medical center in 1006 patients;
recently, New Zealand developed a multivariable risk prediction model, NZRISK, from a large data set of more than 360,000 non-cardiac surgical cases, using eight easily available covariables.
NZRISK has been internally and externally validated in New Zealand and has shown good calibration and discernment in 30-day, 1-year and 2-year mortality models, with a subject operating curve area (AUROC) of 0.92 and McFadden R2 with a statistical value of 0.28, with good resolution.
, NZRISK is not specifically designed for neurosurgery patients, with a calibration slope of 0.72 for neurosurgery patients.
Stephanie Clark of the Department of Anesthesiology and Perioperative Medicine at Auckland City Hospital, New Zealand, developed an easy-to-use neurosurgery patient risk assessment model called the New Zealand Neurosurgery Risk Assessment Tool (The New Zealand Neurosurgery Risk Tool), where both patients and clinicians can participate in data-driven preoperative sharing decisions.
results were published in the September 2020 issue of Neurosurgery.
research method NMDS is data from all inpatient registers recorded by the New Zealand Department of Health.
researchers obtained patient data from NMDS for the development and validation of NZRISK-NEURO over a five-year period from July 1, 2011 to June 30, 2016;
data used for analysis included patients ≥ 18 years of age who under undergo neurosurgery or spinal surgery.
procedure codes in the NMDS database when a patient is admitted to the hospital.
patients who have been excluded from under-the-knee surgery.
neurosurgery patients who underwent more than one neurosurgery during the five-year study were limited to the analysis of the first surgery.
results of the study were mortality rates of 30 days, 1 year and 2 years.
follow-up of patients, blind methods were used to extract the date of death from the birth and death registry.
During the development of NZRISK-NEURO, the predictive effects of five of the eight covariates used in the NZRISK model were analyzed, including asa-PS, surgical emergency, surgical severity, cancer status and age of the Johns Hopkins 5 classification system.
state indicates that active malignancies may affect mortality.
the covariables of surgical surgery to surgical anatomy sites is a predictor of more specific neurosurgery results.
according to the surgical entry and anatomical position, three clinicians divided the type of surgery into intracranial, intracranial, subcranial, spinal and other types.
, five exploratory covariative variables were tested, including gender, diabetes, trauma, race and socioexual status.
New Zealand ethnic groups include New Zealand Europeans, Maori, Pacific, Asians and others.
socio-economic situation is divided into 1 to 10 categories according to the census.
to generate a robust derivative model, each validation covariate requires about 10 events, and each exploratory covariate requires 100 events.
the final model contains 4 validation covariates and 2 exploratory covariates.
data on 18,375 neurosurgery or spinal surgery patients between 2011 and 2016.
modeled multivarivarial risk calculations for each of the main results: 30 days, 1 year and 2 years of mortality.
the predicted values and significant increments of each covariative variable, 8 of the 10 covariative variables initially tested were selected for inclusion in the final model.
predictive factors include age, ASA-PS classification, surgical emergencies (selective or acute), cancer status, anatomy, diabetes, race, and trauma.
predictor of 30-day mortality was ASA-PS level 4 and 5, with an OR value of 13.04 compared to ASA-PS level 1 patients who underwent the same surgery.
the second most important covariation predicting 30-day mortality was anatomy site, where the risk of intracvascular and intracranial surgery was highest (OR values were 3.57 and 3.56, respectively).
"other" anatomical sites, such as celiac sequestration or surgery not confined to an anatomical area, pose a moderate risk (OR=2.40).
30-day mortality rate (OR-1.55) was lower because there were fewer cranial undercover operations in the data set.
the lowest mortality rate from spinal surgery (OR1).
the relative importance of covariable variables changes between the end of 30 days, 1 year and 2 years.
the nature of the surgery was the relevant predictor of death of 30 days, and the OR value of acute surgery and selective surgery was 3.
risk levels of 1 and 2 years of age are not important predictive factors (OR 1.82 and 1.72, respectively).
cancer status is a secondary predictor within 30 days of surgery (OR=1.48), but the highest mortality predictors for 1 and 2 years, OR values of 18.53 and 26.42, respectively.
age and trauma are significant predictive factors in postoperative mortality, but are less important than ASA-PS, cancer conditions, and anatomy sites.
diabetes is common in the data set, affecting 10% of neurosurgery patients, but diabetes is not a major risk factor for mortality (OR values of 1, 1.19 and 1.21 for 30 days, 1 year and 2 years, respectively).
the correlation between race and mortality was weak;
Because there was no statistically significant statistical impact on mortality at 30 days, 1 year or 2 years, 2 of the 10 variables tested were excluded from the final model, including socioeconomic status and surgical severity, the latter effect size of which may be weakened and confused by the overlap between clinical and anatomical site classifications.
final model evaluates performance and versatility through validation of independent datasets.
0.90, 0.91 and 0.91 mortality rates over 30 days, 1 year and 2 years, respectively;
calibrated using a validation dataset, mcFadden R2 statistics are 0.28, 0.37, and 0.41, respectively, and the slope on the calibration chart is 0.93, 0.95, and 0.94, respectively.
Conclusions Research shows that NZRISK-NEURO is a new multivariable risk prediction model that can be used exclusively in neurosurgery and spinal surgery patients;
: The intellectual property rights of the content published by the Brain Medical Exchange's Outside Information, God's Information and Brain Medicine Consulting are owned by the Brain Medical Exchange and the organizers, original authors and other relevant rights persons.
, editing, copying, cutting, recording, etc. without permission.
be licensed for use, the source must also be indicated.
welcome to forward and share.