Risk Adjustment

Which risk-factors are outcomes adjusted for, and which are the most important?

Age (modelled as age plus age-squared), sex, ASA grade, T-stage, N-stage, M-stage, cancer site, mode of admission (from HES/PEDW), number of co-morbidities (Charlson Score from HES/PEDW), and the interaction between age and distant metastases are included in the risk adjustment model. The strongest of the risk-factors that come from NBOCA are age, ASA grade and TNM stage, and it is particularly important that these are complete.

The risk-adjustment model for two-year mortality additionally includes interactions between follow-up period (0-3 months after surgery vs. 3-24 months after surgery) and all of the risk factors. This allows risk factors to have a different effect shortly after surgery and in the longer term. For example, the effect of ASA grade is much larger peri-operatively than in the longer-term, whilst cancer stage has a much larger impact on longer-term than short-term mortality.

The risk-adjustment for 18-month unclosed ileostomy is the same as for 90-day mortality but excludes ‘cancer site’ because it includes only rectal cancer patients.

My trust/network has an unusual distribution of a risk-factor. How will this affect my adjusted outcome?

If your trust/network has unusually low-risk patients compared to all patients nationally, for example they tend to be younger, or have lower ASA grade, or less advanced cancer, the adjusted outcomes for your trust/network will be higher than the observed outcomes. Conversely, if your trust/network has unusually high-risk patients compared to all patients nationally, the adjusted outcomes for your trust/network will be lower than the observed outcomes.

Is TNM staging used in risk-adjustment radiological or histological?

Histological TNM staging is used where it is available; otherwise radiological TNM staging is used. NBOCA no longer collects information about Dukes Staging, therefore is no longer able to update missing M-Stage data. Trusts should ensure M stage is completed, and update from their own record of Dukes’ stage where necessary.

Why are surgical access, surgical urgency and procedure not included in the risk adjustment?

The aim is to adjust for patient factors which reflect frailty of the patient and which cannot be influenced by the provider, so that apart from random variation, any remaining differences in outcomes between providers should reflect differences in quality of care. Providers have influence, to some extent, over the type of surgical procedure, the surgical urgency, and whether laparoscopic techniques are used, and these factors should not, therefore, be adjusted for.

Why does the audit use the pre-2004 NCEPOD classification?

The audit uses the pre-2004 National Confidential Enquiry into Patient Outcomes and Death (NCEPOD) classification of surgical urgency, despite there being an update to this.

The arguments to maintain the pre-2004 NCEPOD definition are that the classification based on this definition correlates strongly with:

  • known risk factors for emergency treatment (age, socio-economic deprivation, and presence
    of co-morbidity)
    • the mode of admission coded in HES
    • the observed 90-day mortality.

Introducing a new classification system for a key characteristic of the surgical procedure would make it impossible to compare outcomes in different audit periods which would in turn make it impossible to monitor trends in outcome over time, which is one of the key functions of the audit.

How is missing information on risk-factors dealt with?

Patients with missing date of surgery are excluded, and multiple imputation, with 10 imputed datasets, is used to fill in any missing information on the risk factors. The method, known as Multiple Imputation using Chained Equations, uses a patient’s other risk-factors to predict their missing information, whilst taking into account the uncertainty due to their missing information. In addition to the data items in the risk-adjustment model, and the outcomes, the following items are used to predict missing risk-factors: surgical urgency, mode of admission according to NBOCA, surgical procedure, number of lymph nodes extracted, number of positive lymph nodes extracted, Index of Multiple Deprivation, length of hospital stay, and days from diagnosis to surgery.

How is patient co-morbidity measured?

The number of co-morbid conditions a patients has is measured using the Charlson Co-morbidity Score, which is calculated using HES/PEDW. A co-morbidity is defined as any hospital admission with one of the following diagnoses in the last year, including the current admission:

  • congestive cardiac failure
    • peripheral vascular disease
    • cerebrovascular disease
    • dementia
    • rheumatological disease
    • liver disease, diabetes
    • hemiplegia/paraplegia
    • AIDS/HIV;

or any of the following diagnoses at a previous hospital admission in the last year:

  • myocardial infarction
    • chronic pulmonary disease
    • chronic renal disease.

The patient does not need to be admitted with the co-morbidity for it to be included. The co-morbidity needs to be included in the patient notes and from there, to make its way into HES, to be included. This can be recorded in the notes at the admission for the bowel cancer resection.

See British Journal of Surgery 2010; 97: 772–781 for more details.