Skip navigation

Calculating the risk of NSAID GI hospital admission

 

NSAIDs are effective medicines for treating arthritis pain and inflammation. They also have adverse effects (and here). One of these that causes much concern is that of gastrointestinal bleeding leading to hospital admission, or even death. One of the problems has been that of predicting who is most at risk of getting these adverse effects. We know that factors like age, NSAID dose, other treatments, and other health problems play a part, but these are difficult things always to bear in mind. A study done in 1991 [1] is some help in trying to work out the risk for an individual, and is an interesting example of prognostic testing.

Reference

JF Fries et al. Nonsteroidal anti-inflammatory drug-associated gastropathy: incidence and risk factor models. American Journal of Medicine 1991 91: 213-222.

Study

The study used people enrolled in ARAMIS (Arthritis, Rheumatism, and Aging Medical Information Study). In five centres 2,747 patients with diagnosed osteoarthritis and rheumatoid arthritis were followed for a total of 9,500 patient years. Each completed a six-monthly detailed heath assessment questionnaire, and there were between one and 13 available for each patient at the time of the study. Information on the patients was available from these questionnaires, from discharge summaries, and from death certificates.

A GI event was defined as one sufficiently serious to result in hospital admission or death. They were counted only when they were the primary basis for hospital admission or death, not if they occurred as part of any other illness. Patients were considered to be taking NSAIDs if this was reported in the questionnaire completed immediately before a GI event occurred (the predictive visit). For controls who did not have an event a random selection of questionnaires was made according to the sequence number for case patients. Thus if 10% of the events occurred after the fifth questionnaire for cases, 10% of the fifth visit questionnaires was selected randomly as controls.

Analysis of demographic, lifestyle, treatment, and comorbid conditions was done by univariate analysis and stepwise logistic regression to find those events that were significantly related to a GI event. From these a model was constructed to predict the risk in an individual.

Results

There was information on 98 cases (13 of whom died) with a defined GI event and on 1,596 controls. The annual rate of GI events while taking an NSAID was 1.6%. Significant differences were observed for a number of factors:

Factor

   
 

With GI event

Without GI event

Age (years, mean)

65

59

Prednisone (% of total)

52

31

NSAID GI adverse effect (% of total)

32

19

NSAID dose (%maximum)

103

91

Disability index (0-3, mean)

1.7

1.4

GI protective drugs (% of total)

41

19

The first five of these remained important in multivariate and with stepwise logistic regression.

Risk model

These results were used to construct a risk model from the five variables that appeared in the logistic regression model. In this model the disability index used in the health assessment questionnaire (0, 1, 2, 3) is roughly equivalent to functional classes 1, 2, 3, and 4 used in the American Rheumatology Association assessment of functional class. NSAID dose is the fraction of the manufacturer's highest recommended dose. Most factors are readily assessable though a simple history. The model is:

GI event scoring table for RA

Factor

Multiply by

Score

Age in years

2

History of previous GI adverse effect (yes 1, no 0)

50

Disability index (0-3) or ARA class minus 1

10

NSAID dose (fraction of maximum)

15

Current prednisone use

40

Total score

 

GI event risk per year on NSAIDs in RA

Risk = (score - 100)/40

A worked example for a 55 year old woman with rheumatoid arthritis taking 150 mg diclofenac daily, who has had Gi adverse effects, has a disability score of 2, and is currently taking prednisone is given below:

GI event scoring table for RA

Factor

Multiply by

Score

Age in years

2

110

History of previous GI adverse effect (yes1, no 0)

50

50

Disability index (0-3) or ARA class minus 1

10

20

NSAID dose (fraction of maximum)

15

15

Current prednisone use

40

40

Total score

235

 

GI event risk (% per year) on NSAIDs in RA

Risk = (score - 100)/40

(235-100)/40

=3.4

So this patient has a calculated annual risk of a GI event of 3.4%.

In the paper the authors calculated the actual risk for patients grouped by 20-unit score categories, and showed a near perfect correlation between actual risk and actual score, and the calculated risk from the scores.

Comment

We have become familiar with using 10-year risks of cardiac events using nomograms that combine age, sex, medical history, blood pressure and cholesterol. The risks with NSAIDs are of a similar magnitude, and a similar proportion (10-20%) die if an event occurs.

This risk calculator for rheumatoid arthritis is welcome. But it lacks testing from an independent data set, and that would be required for us to have real confidence in its use. We would also welcome a similar calculator for osteoarthritis. In the meantime, it stands as an exemplar of the methods that can be used to generate predictive information about patients, predictive information that might change their management.