Skip navigation

Diagnosing prosthetic joint infection

Bandolier has complained about the poor quality of diagnostic test information in the past ( Bandolier 61 ). A new study [1] laying out how to diagnose infection in a prosthetic joint at revision shows, par excellence, how to evaluate a diagnostic test.

In the UK we do over 75,000 hip and knee replacements a year. Some fail, and infection is one of the most serious causes of failure. Infection may present as pain in the joint and loosening of the prosthesis. Treatment may involve removing the joint and treating the infection locally and/or systemically before reimplanting another prosthesis. How best to treat chronic infections will depend on having an accurate and robust means of diagnosing infection.

The study

334 hip and knee revisions over a 17-month period were examined prospectively. The surgical teams were requested to send a standard set of five samples for culture and histology at the time of prosthesis removal. Details of the sampling of tissue and microbiological and histological testing are given in great detail, down to the use of different instruments and fresh scalpel blades to reduce risks of cross-contamination.

The results

Histology results were not available for all patients, so there was information on 297 joints (253 hip, 44 knee) for analysis, 41 of which were infected. Analysis showed (Table 1) that three or more specimens growing the same organism in microbological tests was powerfully related to infection, with a likelihood ratio (LR) of 169.

Microbiology result Likelihood ratio Post-test probability of infection (%)
Three or more specimens positive (same organism) 169 96.4
Two or more specimens positive (same organism) 2.1 25.2
One positive specimen 0.7 10.6
No growth from any specimen 0.2 3.3

Because the likelihood of infection was lower with prostheses which had been in situ for longer periods, the test results were applied to the number of years the prosthesis had been used (Table 2). This showed that even when the prosthesis had been present for more than 10 years and the pre-test probability of infection was as low as 6.6%, three or more positive specimens strongly predicted infection.

    Post-test probability of infection (%)
Years prosthesis in situ Pre-test probability of infection (%) All specimens negative (LR=0.2) One specimen positive (LR=0.7) Two specimens positive (LR=2.1) At least three specimens positive (LR=169)
<2 40 12 33 58 99
2-4 18 4 14 32 97
4-10 8.2 2 6 16 94
>10 6.6 2 5 13 92

Histology with gram staining was shown to be a poor predictor of infection, with a likelihood ratio of 10.


As Mr Punch would say, 'That's the way to do it!'. This is another classic paper demonstrating just how diagnostic strategies should be evaluated. Its value is not just for the diagnosis of infections in hip prostheses, but also as an exemplar for all of us when faced with a new or existing diagnostic test. If it can't meet the standards laid out in this paper, why should we bother to use it?

The other very important aspect of this particular paper is that it is extraordinarily clear about how to reproduce the experimental situation in clinical practice. If, on reading this paper, you decided that you wanted to institute this as practice in your hospital, you could do it. It may be clever, but it is also practical, and full of common sense.


  1. BL Atkins, N Athanasou, JJ Deeks et al. Prospective evaluation of criteria for microbiological diagnosis of prosthetic joint infection at revision arthroplasty. Journal of Clinical Microbiology 1998 36: 2932-2939.

previous or next story in this issue