Objective model using only gender, age and medication list predicts in-hospital morbidity after elective surgery. Blitz JD, Mackersay KS, Miller JC, Kendale SM. British Journal of Anaesthesia 2017;118(4):5444-5550

Presented by: Dr C. Thomas

Background

  • Recognised need for objective, customised risk evaluation tool for elective surgery
  • For patient and physician
  • Aid informed consent
  • Improve safety by identification of high risk patients
  • Current models require physician input / lab data etc.
  • Aim – objective predictor of inpatient post op morbidity
  • Simple to use
  • Easy to include
  • Simple data – age, gender, list of medications
  • Hypothesis:
  • Gender, age and medication list could provide information about post-operative morbidity
  • Certain medications elevate risk
  • Simplified to number of medications / gender / age

Design and Setting

  • Review board approval – patient consent waived as no intervention mandated
  • Restrospective database study
    • Single centre
    • 2 year period
    • Electronic database (Clarity) – access to ICD-9 codes
    • ASA scores from anaesthetist at time (database)
  • Quaternary Care academic Centre – New York City
    • Large inpatient location, ambulatory locations
    • Patients with mod – high access to healthcare
  • Morbidity outcome was in hospital morbidity by
    • Post op complications – presence of any during admission
    • AF, PE, MI, VTE, CCF, Resp Failure, AKI
  • ICD-9 coding limited – excudes:
    • Haemorrhage, sepsis, cardiac arrest
  • Secondary database created:
    • 46 selected medications – presence or absence each patient (on admission)

Subjects

  • 26629 Adult surgical encounters (>18 years)
  • 02% separate patients, 16.98% >1 surgery/patient
  • Anaesthesia – GA / Regional / Neuraxial / Monitored anaesthesia care
  • Exclusions
    • Emergency surgery
    • No ASA score on database

Intervention

  • Developed predictive models for in hospital morbidity based on outcomes above
  • GAMMA – Gender-Age-Medications Morbidity Assessment
    • Morbidity based on gender, age and medications
    • Logistic regretion based on database
  • ASA-M
    • Morbidity using ASA score as independent variable
  • GAMMA-N –GAMMA-Number modification
    • Morbidity solely on gender age and number of medications
  • Binary logistic regression analyses – assessed for discrimination and power by c-statistic (binary outcomes ie yes or no to condition) – >0.8 indicates strong model.
  • Calibration assessed by Brier score (compares actual events with predicted). Score close to 0 suggests accurate.
  • Chi-Square for model significance.
  • Models developed with full data set and validated with k-fold cross validation – 10 folds.

Outcomes

  • Morbidity Risk from gender, age and medications
  • Morbidity Risk from ASA score
  • Morbidity Risk from gender, age and number of medications

Results

  • GAMMA – predicts post operative morbidity with high accuracy (c statistic 0.819, Brier 0.034)
  • ASA similar (c-statistic 0.827, Brier score 0.033)
  • GAMMA-N less predictive (c-statistic 0.795, Brier 0.050)

Conclusions

  • Authors conclude that combination of age, gender and medication list reliably predict post-operative morbidity.
  • Model has increased objectivity, can be used pre-operatively (lab values etc not required, different to models such as PPOSSUM)
  • Limited medical knowledge required therefore could be patient led.

Strengths

  • Large database
  • Authors recognise limitations
  • Easy to access data – on the whole not subjective (except ASA)

Weaknesses

  • Exclusion of haemorrhage, sepsis and cardiac arrest as complications
  • Other outcomes that patients would consider as morbidity? – very limited number of outcomes studied
  • Patient population – excludes limited resource patients – ? therefore not comparable nationally / internationally or patients not on medications for existing disease due to insurance limitations etc therefore risk may be underscored.
  • Limited list of medications included (46) therefore risk may be underscored for patients on less common or new medications etc. How would this be updated with advances in pharmaceuticals?

Implications

  • Difficult to assess from available information
  • If this tool was studied for other populations and proved accurate it could be implemented as a simple risk stratification tool for elective patients but further study would be required.

Potential for impact

  • Development of a patient led tool for risk assessment – patient led care
  • Pre-operative optimization – reduce their score by improving lifestyle etc to reduce medications
  • Risk stratification for allocation of resources? – such as elective joints requiring lowering of BMI before listed for surgery in some areas.