The Trauma Audit and Research Network

PS

The TARN Probability of Survival Model

August 2019

Probability of survival
A probability of survival (PS) is calculated for each injured patient and retained on the TARN database. This allows comparative outcome analyses for hospitals and for other groups of patients to be performed.
 
Early Outcome Prediction Models using TRISS 
In 1984 the Probability of Survival (PS) of each patient was originally calculated from the Revised Trauma Score, Injury Severity Score, age and method of injury (blunt or penetrating). This was known as the TRISS model. There were a number of reasons to develop a European model from this early method:

  1. The Revised Trauma Score incurred a high number of cases with unrecorded data (respiratory rate, systolic blood pressure and Glasgow Coma Scale).
  2. The way that the Injury Severity Score was incorporated into the calculation contradicted some statistical reasoning.
  3. Patients who were transferred to another hospital for further care were excluded.
  4. Patients who were intubated at scene were excluded.
  5. Children were included but not in a statistically acceptable fashion
The first TARN PS model
In 2004 a new PS logistic regression model based on age, gender, Injury Severity Score (ISS) and Glasgow Coma Score (GCS) was launched by TARN. Where GCS was missing, intubation was used instead. Each element in the model carried a weighting derived from retrospective analysis of the TARN database. As the nature of the trauma population changes over time, we recalculated these weightings in 2009 and 2012.

During 2014 we recalculated the coefficients once more and, at the same time updated the model by adding measures to include the comorbidities of patients and a “true 30 day” outcome. This has resulted in 2 case mix standardised outcome (Ws) charts for your hospital.

The coefficients were recalculated again in 2017, the Ws model uses the PS17 values (recalibrated for Ps19).

Why we added comorbidities
For PS to work effectively we must include all characteristics of the injured patients so that we are comparing like with like. In addition to the patient’s age, gender, injuries and level of consciousness, we also need to consider the patient’s state of health. A patient with a severe pre-existing medical condition is different to a patient who was in good health at the time of injury. We handled this comorbidity using a modified version of the Charlson Comorbidity Index, which assigns weightings to certain medical conditions (mCCI). Twenty one groups of comorbidity were created and a weight was allocated to each of these groups. The weights were derived according to the strength of the relationship between the disease group and outcome.

Earlier this year we circulated an email to all hospital staff that informed you of this development and advised that comorbid data would be essential. Data on pre-existing medical conditions (PMC) has also been included in the Accreditation information for this reason. PMC data is essential!

Why we added outcome at 30 days
Outcome (alive or dead) at 30 days from injury has historically been used in the calculation for Ws. However many patients are discharged before this 30 day point. In order to include these patients we need to know whether patients died at or before the 30 day point after leaving hospital. To do this, we now receive information about post-discharge deaths from the Office of National Statistics (ONS) and use this information in one of the calculations of Ws for your hospital. In the future you will receive two Ws charts – one using outcome in hospital and one using the “true 30 day” outcome.. The data linkage is carried out using the patients’ NHS number. We do acknowledge that there are some patients, for example, patients with no fixed abode or who are foreign nationals will not have an NHS number. Excepting this group of patients NHS Number is essential!

The case mix standardised outcome measure Ws
Case mix standardisation uses bands of probability of survival. The bands were revised in Ps17 using an increasingly robust methodology so that there are an equal number of deaths in each band. You will see these in the PS Breakdown table on the TARN website and in your Clinical Reports.


Probability of Survival (Ps19) Model recalibration
The TARN prediction model is calibrated every 2-3 years as a routine exercise.

This is mainly due to the improvement in trauma care and sometimes to a change in the demographics.

The new model, Ps19 is an updated version of Ps17; it uses the same criteria but a change in the age categorisation has been introduced as follows:

< 1yr, 1–10yrs; 11-15yrs; 16–44yrs; 45–54yrs; 55–64yrs; 65–74yrs; 75–84yrs; 85+yrs.

Ps19 has been derived by using information from the TARN database about patients that arrived between April 2017 and March 2019 (inclusive).

Small numbers of patients often affect the accuracy of Ws and therefore the 95% confidence intervals will be large.
Any change in Ws that is encapsulated by these confidence limits means that there is no statistically significant change.

If the numbers of patients submitted are small then you should review the Case ascertainment figures and if <80% try to improve these.

Since 2014 the Ps model has included comorbidity so you must be sure to complete all of this data. A large number of patients with missing PMC data will affect the Ws score.

Detailed Ps19 Model
The Probability of Survival for each patient is calculated using the information in the table below which shows the logistic regression coefficients for patient characteristics (Ps19).

loge is the natural logarithm.
ISS is transformed using fractional polynomial technique for a better fit of the model.
mCCI represents the categorised modified Charlson Comorbidity Index.


b = is defined as the linear combination of the regression coefficients and the values of the corresponding patient’s characteristics (ISS, GCS, modified CCI, age and gender) and the constant e = 2.718282 (the base of Napierian logarithms).


Outcome at 30 days or discharge

Outcome at 30 days via ONS data linkage

Predictors

Coefficient

-2.04374

-1.909581

GCS=3

-4.49297

GCS 4 - 5

-3.34416

GCS 6 - 8

-2.35783

GCS 9 - 12

-1.60302

GCS 13 - 14

-0.52474

GCS 15  (reference)

0.00000

GCS "Intubated"

-3.685728

mCCI Not Known

-0.99161

mCCI 0 (reference)

0.00000

mCCI 1 - 5

-0.51906

mCCI 6 - 10

-0.96600

mCCI > 10

-1.53458

Age < 1

-0.01109

Age 1 - 10

+0.17983

Age 11 - 15

-0.16049

Age 16 - 44 (reference)

0.00000

Age 45 - 54

-0.39462

Age 55 - 64

-0.94640

Age 65 - 74

-1.73794

Age 75 - 85

-2.54999

Age > 85

-3.10822

Gender Male  (reference)

0.00000

Gender Female

-0.1847913

Age &1t; 1 x Female

+0.02051

Age 1 - 10 x Female

-0.26727

Age 11 - 15  x Female

+0.71834

Age 45 - 54 x Female

+0.07896

Age 55 - 64 x Female

+0.12845

Age 65 - 74 x Female

+0.25509

Age 75 - 85 x Female

+0.42293

Age > 85 x Female

+0.49363

Constant

5.67829

 

Predictors

Coefficient

-2.03953

-1.89033

GCS=3

-4.36388

GCS 4 - 5

-3.17872

GCS 6 - 8

-2.11443

GCS 9 - 12

-1.53609

GCS 13 - 14

-0.53992

GCS 15  (reference)

0.00000

GCS "Intubated"

-3.52241

mCCI Not Known

-0.79432

mCCI 0 (reference)

0.00000

mCCI 1 - 5

-0.62056

mCCI 6 - 10

-1.07724

mCCI > 10

-1.74943

Age 0  - 5

0.06055

Age 6 - 10

0.47385

Age 11 - 15

-0.08743

Age 16 - 44 (reference)

0.00000

Age 45 - 54

-0.33766

Age 55 - 64

-0.90488

Age 65 - 74

-1.70869

Age > 75

-2.85978

Gender Male  (reference)

0.00000

Gender Female

-0.18650

Age 0  - 5 x Female

-0.24235

Age 6 - 10 x Female

0.02104

Age 11 - 15  x Female

0.59432

Age 45 - 54 x Female

-0.10280

Age 55 - 64 x Female

0.22073

Age 65 - 74 x Female

0.25692

Age > 75 x Female

0.43942

Constant

5.49085