When dipstick urinalyses for nitrite and leukocyte esterase are simultaneously negative it is unlikely that the urine culture will show growth of potentially pathogenic bacteria. There are no clinically relevant differences between visual and analyzer readings of the dipsticks.
Methodological aspects
In this study we obtained a urine specimen from 55% (651/1187) of all individuals registered at the nursing homes. This may appear low but approximates previously published studies in nursing homes for elderly [
4]. The main reason for nonparticipation in this study was substantial urinary incontinence. Most of these individuals also had dementia. The only possibility of obtaining a urine specimen from these individuals would have been by catheter. This is not routine for clinical practice for elderly at nursing homes and would, therefore, not have been representative of clinical practice. Furthermore, this would have been unethical. Individuals with an indwelling urinary catheter were excluded as they always become colonized by bacteria sometimes of different species compared to those without [
1]. Only 12% (96/791) actively refused participation which we considered acceptable.
The study by Juthani-Mehta et al [
15] presenting confidence intervals for PPV and NPV included only patients with symptoms of suspected UTI. Specific symptoms were dysuria (7%), change in voiding pattern (6%) or fever (12%) and unspecific symptoms were change in mental status (40%), behavior (20%), character of urine (17%), and evaluation for other infection (7%), family or patient request (7%), etc. However, it is unclear which clinical features or events are relevant in bacteriuria [
1,
29]. Thus, while Juthani-Mehta et al attempt to estimate dipstick analysis ability to predict UTI, this study focused on evaluating dipstick ability to predict bacteriuria, not UTI. Prevalence of bacteriuria among asymptomatic residents in nursing homes for elderly is high [
1,
4,
5] and similar to the prevalence found by Juthani-Mehta et al (40%) and in this study (32%). Since PPV and NPV for dipstick analysis depend on prevalence of bacteriuria there should be no major differences between evaluating dipstick analysis for symptomatic or asymptomatic individuals.
Only 4.0% (26/651) had ongoing antibiotic treatment thus no urinary bacteria growth and negative nitrite dipstick may have been expected. However, leukocyte esterase dipstick may remain positive for some time. Thus, these patients were more likely to influence test results of leukocyte esterase rather than nitrite dipstick. Due to the low prevalence of ongoing antibiotic treatment this effect was considered low.
Procedures allowing presence of a few specific symptoms or outcomes of prior dipstick testing influence the decision of cut-off levels for CFU in urine culture may enhance the diagnostic procedure [
30]. These procedures are very common in microbiologic laboratories in Sweden. Thus, the present procedure for urine culture was used without modification to be representative of ordinary clinical practice.
Dipstick urinalysis was performed by non-laboratory personnel in this study. If the analysis had been performed by laboratory personnel, results might have differed slightly. On the other hand, these bedside tests are usually performed by non-laboratory personnel in clinical practice at nursing homes for the elderly. Thus, this study represented ordinary clinical practice.
The results of this study can be considered generalisable in developed countries when evaluating urine dipstick analysis for elderly individuals at nursing homes performed in ordinary clinical practice.
The NPV for nitrite to predict absence of E. faecalis in a urine culture is higher than for E. coli despite E. faecalis being a poor converter of nitrate to nitrite. The most likely explanation being the prevalence of E. faecalis is very low (2.6%) resulting in a high NPV even if sensitivity and specificity are low.
The kappa coefficient for agreement between visual and analyzer readings was lower for leukocyte esterase dipsticks than for nitrite dipsticks. This is logical whereby leukocyte esterase dipsticks have several color blocks while nitrite dipstick has only a binary outcome. The more possible outcomes the lower the kappa value becomes.
The effect of Simpson's paradox when evaluating dipsticks
It may seem peculiar that PPV for any bacteria is higher than PPV for a single bacterium [see Additional file
1] [see Additional file
2] [see Additional file
3] [see Additional file
4]. One explanation is prevalence of bacteria in the gold standard is higher when the focus is on "any bacteria" compared to a specific bacterium subsequently decreasing the probability of a false positive dipstick. The reverse was seen for NPV.
Another way to explain this phenomenon is the well known Simpson's paradox. Several potentially pathogenic bacteria differ in their ability to reduce nitrate to nitrite. Similarly, different bacteria are likely to show a varying ability to provoke pyuria. This difference is a confounding factor and as the prevalence of the different types of bacteria varies considerably the size of these groups vary. This phenomenon has been previously explained as Yule-Simpson's effect, a statistical paradox in which the outcome of several groups is changed when groups are combined [
24‐
26].
The conclusion that when dipstick urinalyses for nitrite and leukocyte esterase are simultaneously negative it is unlikely that the urine culture will show growth of potentially pathogenic bacteria is based on "any bacteria" [see Additional file
4]. Furthermore, NPV for each single bacterium is higher thus the conclusion seems valid. If one should decide that positive nitrite dipstick can rule in bacteriuria, as a previously published metaanalysis did [
16], we find that PPV for "any bacteria" differs very much from PPV for the single bacteria [see Additional file
2]. Thus, the conclusion that positive nitrite dipstick can rule in bacteriuria seems unjustified.