Urine in the bladder is normally sterile, or close to it. A urinary tract infection is what happens when bacteria — most often Escherichia coli from the person's own bowel — travel up the urethra, colonise the bladder and provoke inflammation.
The anatomy explains almost all of the epidemiology. The female urethra is short and its opening sits close to the anus, so the journey is short and the source is nearby. That is why UTI is overwhelmingly a disease of women: 10.8 percent of US women aged 18 and over reported at least one presumed UTI in the past 12 months, an estimated 11.3 million women treated with antibiotics in 1995 at a cost of about $1.6 billion, and by age 24 roughly one-third of women will have had at least one physician-diagnosed, prescription-treated UTI [1].
Where the bacteria stop determines how serious it is.
- Cystitis is infection of the bladder — the lower tract. Burning on passing urine, urgency, frequency, sometimes blood. It is unpleasant and disruptive, and it is not dangerous.
- Pyelonephritis is infection that has ascended to the kidney — the upper tract. Fever, flank pain, nausea, systemic illness. It can cause sepsis and it can damage the kidney. It is a different clinical problem requiring longer, broader treatment [2].
Most uncomplicated cystitis is easy to treat, and a short course of the right antibiotic resolves it. So a review of UTI could be very short. It is not, because the two things that actually make this disease difficult are not the initial infection:
Recurrence. In a cohort of women followed after a first UTI, 26.6 percent had a culture-confirmed recurrence within six months [3]. Among infection-prone women followed a median of nine years off prophylaxis, infections ran at 2.6 per patient-year with wide variation between individuals (0.3 to 7.6 per year), and episodes were strikingly clustered [4]. Susceptibility is partly inherited: a UTI history in a mother, sister or daughter raised the odds of recurrent cystitis 3.1-fold with one affected relative and 5.0-fold with two or more [5] [6].
Resistance. UTI is one of the largest single drivers of community antibiotic prescribing, and the antibiotics that work are being lost. That is the subject of the centerpiece.
Centerpiece: empiric therapy works only while the bug stays susceptible
Uncomplicated cystitis is treated empirically — the antibiotic is chosen and started before anyone knows what grew, and often no culture is sent at all. That is defensible only because the causative organisms are predictable: in an international survey of 4,734 women, pathogens were found in 3,278 and E. coli accounted for 77.0 percent of isolates [7]; in a US series of 4,342 isolates, E. coli and Staphylococcus saprophyticus together were 90 percent [8].
But predictable identity is not the same as predictable susceptibility. The left panel is the resistance landscape from that international survey: in E. coli, resistance was 29.8 percent to ampicillin, 29.1 percent to sulfamethoxazole, 14.8 percent to trimethoprim and 14.1 percent to trimethoprim–sulfamethoxazole, while nitrofurantoin, fosfomycin, mecillinam, ciprofloxacin, gentamicin, cefadroxil and co-amoxiclav were all below 3 percent [7]. In the US series the same split appeared over time: TMP-SMX resistance in E. coli rose from more than 9 percent in 1992 to more than 18 percent in 1996 — a doubling in five years — while nitrofurantoin, gentamicin and ciprofloxacin stayed at 0–2 percent and did not change [8] [9] [10].
Why the 20 percent line matters is the right panel, and it comes from a trial that answered the question directly. Premenopausal women with symptomatic lower UTI were all given TMP-SMX before susceptibility was known. 71 percent of cultures grew susceptible organisms and 29 percent grew resistant ones. Microbiological cure was 86 percent in the susceptible group and 42 percent in the resistant group [11].
Those two numbers define a model. Expected cure with empiric therapy is a mixture of the two, weighted by local resistance:
expected cure = (1 − R) × 86% + R × 42%
The check the arithmetic was never given. The trial prints 71/29 and prints 86 and 42, and never multiplies them out to say what its own cohort achieved overall. It is 73.2 percent. And the model has a slope: every 1 percentage point of local resistance costs 0.44 points of expected cure, because that is the gap between curing 86 percent and curing 42 percent.
Feeding the guideline threshold through it gives the number that justifies the rule. At exactly 20 percent resistance, expected cure is 77.2 percent — 8.8 points below what the same drug achieves against a susceptible organism.
And a less comfortable derived result. Expected cure falls below 80 percent at a resistance of just 13.6 percent — below the conventional cutoff. By the time an agent is formally retired at 20 percent, it has already spent time performing worse than four cures in five. The 20 percent line is a pragmatic compromise, not a point at which something suddenly breaks.
The contrast that explains current first-line choices. At nitrofurantoin's reported ceiling of 3 percent the model loses 1.3 points of cure; at TMP-SMX's 14.1 percent it loses 6.2. That few-point difference is the whole reason nitrofurantoin, fosfomycin and pivmecillinam are recommended first-line for cystitis while fluoroquinolones are reserved and TMP-SMX is conditional on local rates [12] [2].
Three honest limits. The line is a model, not an observation: it assumes the 86 and 42 percent cure rates hold at every resistance prevalence, which the trial could not test — the five marked points are grounded and the line between them is arithmetic. The survey figures are pooled across 17 countries, and the paper's own finding is that they vary enormously — lowest in the Nordic countries and Austria, highest in Portugal and Spain, where ciprofloxacin resistance reached 14.7 percent [7]. And these data are from the 1990s and 2000s; the point is the structure of the argument, not the current value of R in any particular place.
The clinical consequences are documented directly, not just modelled. Women recently exposed to TMP-SMX were more than 16 times as likely to carry a resistant isolate, and women infected with a resistant isolate and treated with TMP-SMX were more than 17 times as likely to fail [9]. In a head-to-head trial, clinical cure with TMP-SMX was 84 percent when the organism was susceptible against 41 percent when it was not [13] — an independent replication of the same two numbers the model is built from.
The teaching point. Empiric therapy is a bet on the local resistance rate. Every point of resistance costs 0.44 points of cure, which is why first-line choice must follow local susceptibility data and why protecting the agents that still work — by not treating what does not need treating, and by using short courses — is not an abstraction.
One sobering caveat about that last claim. A regional intervention that cut trimethoprim use by 85 percent produced only a marginal effect on trimethoprim resistance, with no detectable fitness cost to the resistant strains and strong co-selection by other antibiotics [14]. Stewardship is far better at preventing resistance than at reversing it.
Pillar 1: measurement and diagnosis
In typical uncomplicated cystitis, the diagnosis is usually clinical
A healthy non-pregnant woman with new burning, urgency and frequency, and no vaginal discharge, has a high enough probability of cystitis that empiric treatment without a culture is standard [2] [12] [15].
The dipstick supports that judgement without settling it. A meta-analysis of 70 publications found nitrite accuracy high in pregnant women (diagnostic odds ratio 165) and the elderly (108), leukocyte-esterase accuracy high in urology patients (276) and most sensitive in family medicine (86 percent), with combining both tests raising sensitivity to 90 percent in family practice. Its conclusion is the practical one: the dipstick is useful mainly to rule infection out when both nitrite and leukocyte esterase are negative [16] [17]. Pyuria — white cells in the urine — has its own long-standing relationship to bacteriuria [18].
And the negative case matters more than clinicians usually assume. In an emergency-department study of 264 women treated for UTI, only 48 percent had a positive urine culture; 23 percent had a positive sexually transmitted infection test, of whom 37 percent went untreated for it — and 64 percent of those had been diagnosed with a UTI instead. Ninety-two percent had an abnormal urinalysis, but its positive predictive value was only 41 percent [19]. Overdiagnosis of UTI and missed STI travel together.
When to culture, and how to read one
Culture with susceptibilities is indicated for complicated infection, recurrence, pyelonephritis, pregnancy, and treatment failure [2] [20].
Reading it requires knowing which organisms in a voided sample actually came from the bladder. In premenopausal women with acute cystitis, midstream and catheter specimens were compared directly: E. coli in midstream urine predicted bladder bacteriuria even at very low counts, with a positive predictive value of 93 percent at 10² colony-forming units per millilitre, while enterococci and group B streptococci were not predictive at any colony count — and in 61 percent of episodes where those organisms appeared, catheter urine grew E. coli [21]. Growth of E. coli means something at low counts; growth of the other two often means contamination.
Asymptomatic bacteriuria, and the stewardship point that follows
Bacteria in the urine without symptoms is a distinct condition, and in most people it should not be treated. It is common: prevalence was 5 to 6 percent in sexually active young women, rarely persistent, and a strong predictor of subsequent symptomatic UTI — symptomatic infection followed within a week of 8 percent of positive asymptomatic cultures against 1 percent of negative ones [22]. It is commoner still in older women [23] [24] and in diabetes [25] [26].
The trials are consistent and the direction is uncomfortable. In diabetic women with asymptomatic bacteriuria, antimicrobial treatment cleared the bacteriuria — 20 percent still bacteriuric at four weeks against 78 percent on placebo — but over a mean 27 months, 40 percent of the placebo group and 42 percent of the treated group had at least one symptomatic UTI, with no difference in time to first episode, pyelonephritis or hospitalisation [27]. In elderly women, asymptomatic bacteriuria was not an independent risk factor for mortality, and treatment did not lower it (13.8 versus 15.1 deaths per 100,000 resident-days) even though it raised cure from 15.6 to 82.9 percent [28] [29]. And in young women with recurrent UTI, the group whose asymptomatic bacteriuria was left alone had markedly fewer symptomatic recurrences — 13.1 percent against 46.8 percent at last follow-up — leading the authors to suggest the bacteriuria may be protective [30].
Sterilising the urine is achievable. It just does not help, and may hurt.
The exceptions are specific and well defined: screen and treat in pregnancy, and before urologic procedures that breach the mucosa such as transurethral resection of the prostate [31] [32] [33] [34]. The pregnancy case is strong: antibiotic treatment reduced pyelonephritis (risk ratio 0.23, 0.13–0.41 across 11 studies and 1,932 women), low birthweight (0.64) and preterm birth (0.27) [35] [36] [37] [38] [39] [40], with the Cochrane authors noting all 14 studies were at high or unclear risk of bias. Diagnostic accuracy in pregnancy has been quantified — a positive dipslide gives a likelihood ratio of 225 while nitrite-plus-leukocyte-esterase dipsticks give only 6.95 [41]. The US Preventive Services Task Force reached the same split: screen in pregnancy, do not screen non-pregnant women with diabetes [42] [43].
Why the rule is broken so often is itself studied. In institutionalised elderly people, ordering of urine cultures and prescribing for asymptomatic bacteriuria were driven by non-urinary symptoms and signs — vague changes in status communicated by nursing staff — which physicians felt justified antibiotics despite the absence of evidence [44] [45].
Uncomplicated versus complicated
The distinction governs everything downstream. Uncomplicated means a healthy, non-pregnant woman with a structurally normal urinary tract [2] [20]. Complicated covers men, pregnancy, catheters, obstruction or stones, immunocompromise and structural abnormality — where organisms are more varied and resistant, courses are longer, and the underlying problem needs addressing.
Catheters deserve their own note because catheter-associated bacteriuria is largely inevitable and largely should not be treated. It has formal guidelines [46] [47], its complications in long-term catheterisation are described [48], pyuria does not distinguish infection from colonisation in catheterised patients [49], and the antimicrobial-impregnated and silver-coated catheters marketed to prevent it have been systematically reviewed with modest results [50] [51] [52] [53] [54]. The effective intervention is removing the catheter.
Pillar 2: treatment and prevention
Uncomplicated cystitis: short courses of the right agent
First-line agents are nitrofurantoin, trimethoprim–sulfamethoxazole where local resistance is low, fosfomycin and pivmecillinam; fluoroquinolones should not be routine first-line for cystitis, although they are first-line empiric therapy for pyelonephritis [12] [2].
Nitrofurantoin has good clinical and microbiological efficacy with clinical cure of 79 to 92 percent, equivalence to TMP-SMX, ciprofloxacin and amoxicillin when given for 5 or 7 days, and mild, mostly gastrointestinal toxicity — but its efficacy drops to 61–70 percent if given for only 3 days, and acquisition of resistance to it remains rare [55]. A 5-day course was clinically and microbiologically equivalent to 3 days of TMP-SMX [13], and against placebo it produced a number needed to treat of 4.4 at three days and 2.7 at seven [56].
Fosfomycin as a single dose showed no difference from comparators in clinical success across 10 randomised trials and 1,657 patients (risk ratio 1.00), with significantly fewer adverse events in pregnant women [57] [58] [59], and it retains activity against ESBL-producing organisms [60] [61].
Pivmecillinam was superior to placebo at every regimen tested, with seven-day courses outperforming three-day (bacteriological cure 93–94 versus 84 percent) [62].
The duration question has a general shape: shorter is usually enough, but not for every drug — nitrofurantoin and pivmecillinam both lose efficacy at three days, while TMP-SMX and fosfomycin do not [55] [62] [57] [63].
Pyelonephritis
Upper-tract infection needs longer and broader treatment, and sometimes admission. In a randomised comparison, 7 days of ciprofloxacin achieved bacteriological cure in 99 percent and clinical cure in 96 percent, against 89 and 83 percent for 14 days of TMP-SMX — with E. coli, causing over 90 percent of infections, more often resistant to TMP-SMX [64]. Seven days of ciprofloxacin has since been compared against 14 [65], and five days of high-dose levofloxacin against seven of ciprofloxacin [66]. Pyelonephritis in pregnancy is a distinct and more dangerous entity [67] [68] [69], and risk factors in otherwise healthy women have been characterised [70] [71] [72].
Recurrent UTI
Vaginal oestrogen is the standout intervention in postmenopausal women, and it works by restoring the ecology rather than by killing anything. In a controlled trial of intravaginal estriol, UTI incidence fell from 5.9 to 0.5 episodes per patient-year — nearly a twelvefold reduction. Lactobacilli, absent from every vaginal culture at baseline, reappeared in 61 percent of estriol recipients and none of the placebo group; vaginal pH fell from 5.5 to 3.8; and colonisation with Enterobacteriaceae fell from 67 to 31 percent while remaining unchanged on placebo [73] [74].
That mechanism is corroborated independently. Vaginal E. coli colonisation was significantly more frequent in women with recurrent UTI than controls (35 versus 11 percent) and in women lacking hydrogen-peroxide-producing lactobacilli (odds ratio 4.0), with spermicide use associated with a 12.5-fold odds of vaginal E. coli colonisation [75] [76]. A Lactobacillus crispatus vaginal probiotic reduced recurrence from 27 to 15 percent in a phase 2 trial (relative risk 0.5, 0.2–1.2) — not statistically significant overall, but significant in women who achieved high-level colonisation [77] [78].
Behavioural measures follow from the same epidemiology: diaphragm-plus-spermicide use is a repeatedly identified risk factor [79] [80] [22], as is sexual intercourse [79] [81].
Antibiotic prophylaxis works and has a cost. Postcoital TMP-SMX was highly effective against placebo in women whose infections related to intercourse [82], and continuous prophylaxis prevented cystitis, asymptomatic bacteriuria and pyelonephritis even over five years — while the proportion of infecting strains resistant in vitro rose [4] [83]. Patient-initiated self-treatment is a way to get the benefit with less exposure: among 88 women self-diagnosing 172 UTIs, 84 percent had a uropathogen on culture, with clinical and microbiological cure of 92 and 96 percent of confirmed episodes and no serious adverse events [84].
On cranberry, the honest answer is that it does not work for prevention. The mechanistic story is real — proanthocyanidins with anti-adhesion activity rather than urinary acidification [85] — and early studies showed reductions in bacteriuria and pyuria [86]. But the randomised prevention trials are negative: in college women with an acute UTI, twice-daily cranberry juice produced no decrease in six-month recurrence, with the cranberry group slightly higher (20.0 versus 14.0 percent) [87]; a cranberry-lingonberry and Lactobacillus trial [88], a naturopathic cranberry trial [89] and a trial in pregnancy [90] add little. The mechanism being plausible has not made the intervention effective.
Treating the underlying complicating factor — obstruction, stones, an enlarged prostate — is what converts recurrent infection from a prophylaxis problem into a solved one, and Proteus mirabilis urease links infection and stone formation directly [91].
The biology, briefly, because it explains the recurrence
Uropathogenic E. coli is not a passive contaminant. Type 1 pili bind uroplakins on the bladder surface; attachment triggers exfoliation of bladder epithelial cells by an apoptosis-like mechanism as an innate defence — and the bacteria evade that clearance by invading into the epithelium [92] [93]. Virulence factors and pathogenicity islands distinguish uropathogenic strains from commensals [94] [95] [96] [97], and host defences including neutrophil recruitment are genetically variable [98] [99] [6]. An organism that can hide inside the bladder wall is an organism that can come back.
Pillar 3: progress
Non-antibiotic prevention is the most useful direction, and vaginal oestrogen is the one with the strongest evidence in this substrate [73], with vaginal probiotics behind it [77].
Two gaps must be stated. The task specification lists methenamine hippurate and UTI vaccines as progress areas. This substrate contains one methenamine study, in patients with neurogenic bladder from 1984 [100] — no ALTAR-type non-inferiority trial — and no vaccine trial at all; vaccine strategies appear only as a topic heading in a review [93]. This review therefore does not characterise either. Readers should go elsewhere for both.
Rapid diagnostics and susceptibility testing are the natural answer to the centerpiece: if you knew the susceptibility at the point of care, the mixture model would collapse and empiric therapy would stop being a bet. The substrate supports the need rather than a product — the diagnostic-accuracy literature on dipsticks [16] and cultures [21] defines what has to be improved on, and whole-genome prediction of resistance has been demonstrated in another organism [101].
Resistance surveillance is what makes local first-line choice possible at all [7] [102] [8] [103] [104], and it now has to track more than susceptibility percentages. The globally disseminated ST131 clone producing CTX-M-15 was described across continents with highly homogeneous virulence genotypes [105] [106], ESBL-producing urinary E. coli has its own risk-factor profile — nursing-home residence, diabetes, recurrent UTI, high comorbidity — and is associated with more symptoms and longer admission [107] [108] [109]. Resistance in UTI is increasingly the spread of particular successful clones, not just gradual drift.
Reducing overtreatment of asymptomatic bacteriuria is the highest-yield stewardship target because the evidence is unambiguous and the practice is entrenched [31] [32] [44] [45] — and, as the trimethoprim-reduction experiment showed, the resistance you prevent is worth far more than the resistance you can later reverse [14].
Dig deeper in lmmol
Kidney stones and UTI interact in both directions: stones and obstruction are among the defining features that make an infection complicated [2], and Proteus mirabilis urease contributes directly to urolithiasis as well as to persistence and pyelonephritis [91] — an infection that makes stones that then shelter the infection. Chronic kidney disease is the organ at stake when infection ascends; long-term renal outcomes after childhood UTI have been followed directly [110], and pyelonephritis is the mechanism by which a bladder infection becomes a kidney problem [2]. Benign prostatic hyperplasia is the commonest cause of the obstruction that makes UTI in men complicated by definition, and prostatic disease sits alongside it in the differential [111]. Diabetes and glycemic control raises both asymptomatic bacteriuria and symptomatic UTI risk [25] [112] [26] — and is also where the clearest negative trial of treating asymptomatic bacteriuria was done [27]. Pneumonia is the other great driver of community antibiotic prescribing and shares this review's central tension exactly: empiric therapy chosen before the organism is known, judged against local resistance, with duration as the lever stewardship can actually pull. Sepsis is where untreated upper-tract infection ends, and bacteriuria in long-term catheterised patients has documented progression to fever, bacteraemia and death [48]. The full collection is at health.