Sixteen studies, with a total of 3,711 participants (range 54 to 431), were included in the review. Spectrum bias was a potential problem in four studies, which either included patients with both complicated and uncomplicated urinary tract infection or did not report whether they included such patients. Partial verification bias was a potential problem in two further studies, and the presence of uninterpretable test results and blinding were generally poorly reported.
Using a threshold of 102 or greater CFU/ml for the diagnosis of urinary tract infection, the sensitivity estimates for signs and symptoms ranged from 10% (95% CI 4 to 21) for fever (based on seven studies) to 80% (95% CI 74 to 86) for dysuria and 88% (95% CI 83 to 92) for frequency (based on 14 and 13 studies, respectively). Specificity estimates ranged from 20% (95% CI 14 to 28) for frequency to 92% (95% CI 83 to 97%) for fever.
Patterns were similar using a threshold of 103 or greater CFU/ml for the diagnosis of urinary tract infection. The sensitivity estimates for signs and symptoms ranged from 12% (95% CI 5 to 26) for fever (based on six studies) to 88% (95% CI 82 to 92) for frequency (based on 11 studies). Specificity estimates ranged from 21% (95% CI 14 to 31) for frequency to 91% (95% CI 80 to 97) for fever.
The same pattern persisted when a diagnostic threshold of 105 CFU/ml was used. Frequency had the highest sensitivity (90%, 95% CI 85 to 94) and lowest specificity (17%, 95% CI 11 to 26), six studies. Fever had the highest specificity (89%, 95% CI 75 to 95) and lowest sensitivity 10% (95% CI 4 to 23), four studies.
Estimates of positive and negative likelihood ratios were also reported.
The authors used data on the accuracy of urine dipstick tests, from another systematic review, to produce combined estimates of post-test probability of urinary tract infection (signs and symptoms plus dipstick results).