Prediction is hard and prediction is even harder for postoperative infections

Deep infection after a major orthopaedic surgery is a devastating complication. A prediction model with excellent predictive performance for deep infections would be extremely valuable for example in joint replacement surgery. The problem is that a deep infection is (fortunely) very rare event. Overall infection rate is close to 1% in the modern total knee or hip replacement surgery. This makes development of valid prediction models very hard.

Several authors have proposed prediction model for deep infection after total joint replacement. A closer look to these models will provide good insight why infection prediction remains as a very challenging job.

Everhart et al. developed a risk model based on 6789 TJA patients which is a very nice sample size for this sort of study. Their scoring system is shown below:

If our hypothetical patient is a smoker and a long-term insulin user and he/she is referred for a total hip replacement surgery, the absolute estimated risk of infection is 2.8%. Even though this combination of risk factors seems very unpleasant for major surgery, the estimated probability of an infection is still relatively low and might not have that large influence on the patient preferences. If our patient was also morbidly obese, the estimated risk increases to 6.6%. Again, this might not seem high enough for the patient to avoid a surgery although from the clinical perspective this patient is everything but a candidate for a THA. The point here is to say that even if the patient is a smoker, has insulin-dependent diabetes and is morbidly obese, the absolute risk may seem acceptable for a treatment of debilitating joint pain.

Another well known study was published by Tan et al. Their scoring system is as follows:

Based on this model, the estimated risk of an infection after primary TKA surgery is 2.19% for a male patient who is morbidly obese (BMI 42) and who is a smoker. Again this absolute estimated risk is surprisingly low considering that the patient is far from optimal candidate TKA surgery.

Higher estimated absolute risks in these models are seen if patient has had a septicemia, or prior periprosthetic infection, has primary bone cancer or patient has both prior drug abuse and a renal disease. Such patients are, however, extremely rarely referred for THA or TKA surgery. This means that the proportion of patients with truly high absolute risk (>10%) of infection is very low resulting to very low overall prevalence of infections. This on the other hand means that an infection is a rare event after TKA or THA and prediction becomes quite hard as said in the first paragraph. Advanced methods do not seem to predict an infection much better compared to the statistical methods and very high uncertainty remains still after the most important question in TKA and THA surgery: which of my patients will develop a postoperative deep infection?

Leave a Reply

Your email address will not be published. Required fields are marked *