How Empiric Antibiotics Are Chosen Before Culture Results

Empiric antibiotics are drugs given before doctors know exactly which bacterium is causing an infection, chosen on the basis of the most likely culprits and local resistance patterns rather than lab-confirmed results. In life-threatening infections like sepsis, every hour of delay in starting antibiotics raises the odds of dying, so waiting for culture results is often not an option. Yet the practice creates a genuine tension: going broad enough to cover dangerous bacteria means many patients receive drugs they never needed, with real consequences for their kidneys, their gut, and the wider problem of antibiotic resistance.

Why Starting Early Matters So Much in Sepsis

The strongest case for empiric antibiotics comes from sepsis, where a patient’s immune response to infection spirals out of control and can damage organs within hours. A large retrospective study of over 35,000 patients found that each hour of delay in giving antibiotics raised in-hospital mortality by about 0.3% in sepsis overall, 0.4% in severe sepsis, and a striking 1.8% in septic shock.¹ Those numbers may sound small per hour, but they accumulate quickly: a patient in septic shock whose antibiotics are delayed by three or four hours faces a meaningfully higher chance of death.

A 2024 systematic review and meta-analysis confirmed this pattern across multiple studies. Compared to receiving antibiotics within the first hour, each additional hour of delay raised the odds of dying in-hospital by about 4%, and receiving antibiotics after the first hour altogether increased those odds by roughly 20%.² This is the core logic behind empiric prescribing in emergencies: you treat now with your best guess and refine later.

What “Best Guess” Actually Looks Like

In practice, empiric therapy for suspected sepsis tends to be broad. A study of nearly 900,000 U.S. hospital admissions for suspected community-onset sepsis found that about 42% of patients received an agent targeting MRSA (methicillin-resistant Staphylococcus aureus), about 57% received an anti-pseudomonal beta-lactam, and roughly 35% received both. The most commonly prescribed drugs were vancomycin, ceftriaxone, piperacillin-tazobactam, and cefepime.³

Here is the uncomfortable finding from that same study: among patients who received these broad-spectrum drugs, over 90% turned out not to have the resistant organisms the drugs were targeting. That fraction actually grew over time, reaching nearly 92% by 2021. In other words, the vast majority of patients given heavy-duty antibiotics for MRSA or Pseudomonas did not have MRSA or Pseudomonas.⁴ This is the central paradox: doctors cannot afford to guess wrong when lives are at stake, so they go wide, but going wide means most patients are overtreated.

Choosing Smarter With Local Data

One of the most practical tools for improving empiric prescribing is the antibiogram, a facility-specific summary of which local bacteria are susceptible to which antibiotics. Hospitals compile these from their own culture and sensitivity data, sometimes broken down by ward, patient population, or specimen type. The idea is straightforward: if your hospital’s E. coli strains are resistant to a particular drug 40% of the time, you probably should not reach for it first when treating a likely E. coli infection.

A prospective study at a tertiary care hospital in India found that creating an antibiogram toolkit and making it available at the point of care helped clinicians select more appropriate empiric regimens and reduce irrational antibiotic use.⁵ The concept is simple, but many hospitals still do not make this information easily accessible to prescribers at the bedside. Where antibiograms exist and are actively used, they function as a local cheat sheet that narrows the guesswork considerably.

De-Escalation After Culture Results

The flip side of starting broad is narrowing down once you know what you are dealing with. De-escalation means switching from a broad-spectrum antibiotic to a narrower one, or stopping unnecessary drugs altogether, once culture results identify the pathogen and its sensitivities. This step is crucial for limiting harm, but it often happens late or not at all.

A large study of adults hospitalized for community-onset sepsis found that de-escalating from anti-MRSA or anti-pseudomonal therapy did not increase 90-day mortality compared to continuing broad-spectrum drugs. Patients whose therapy was narrowed had fewer total days of antibiotics and shorter hospital stays.⁶ Similarly, a study of patients with severe community-acquired pneumonia found no association between de-escalation within the first several days and increased mortality, though de-escalated patients received fewer days of intravenous antibiotics.⁷

The evidence is reassuring but comes with a caveat worth noting. A systematic review covering bloodstream infections and pneumonia found that observational studies consistently associated de-escalation with lower mortality, but randomized trials did not confirm this benefit and even showed a non-significant trend toward increased mortality in the de-escalation arm.⁸ The likely explanation is selection bias: in observational studies, healthier patients get de-escalated while sicker ones stay on broad-spectrum drugs. The takeaway is that de-escalation appears safe for most patients, but doctors still need clinical judgment rather than a blanket policy.

The Real Costs of Going Too Broad

Empiric antibiotics are not free interventions that merely carry the risk of being unnecessary. They have concrete harms.

One well-documented example involves the kidney. Vancomycin combined with piperacillin-tazobactam, one of the most common empiric pairings in U.S. hospitals, carries a significantly higher risk of acute kidney injury than vancomycin paired with other anti-pseudomonal drugs. A network meta-analysis found the combination roughly doubled the odds of kidney injury compared to vancomycin plus cefepime, and roughly doubled the risk of severe kidney injury as well.⁹ A matched cohort study put numbers on it more starkly: kidney injury occurred in about 21% of patients on vancomycin plus piperacillin-tazobactam versus about 13% on vancomycin plus cefepime.¹⁰ This is not a rare complication. It affects a large fraction of patients receiving a drug combination that many hospitals still use as their default empiric regimen.

Broad-spectrum antibiotics also disrupt the gut microbiome, which is central to the risk of Clostridioides difficile (C. diff) infection, a dangerous cause of severe diarrhea and colon inflammation.¹¹ Among broad-spectrum agents, carbapenems appear to carry the highest risk: one study found that patients receiving carbapenems had roughly a fivefold increased risk of C. diff infection within 90 days compared to those on other antibiotics.¹² These gut complications can extend a hospital stay, require additional treatment, and sometimes kill. They are a direct consequence of empiric antibiotic choices.

Procalcitonin and Knowing When to Stop

One of the hardest decisions in empiric therapy is not which drug to start but when to stop. Blood cultures take time, infections can be hard to confirm, and clinicians often err on the side of continuing treatment. Procalcitonin, a blood marker that rises during bacterial infection and falls as it resolves, has become an increasingly used tool to guide those decisions.

A meta-analysis of 16 randomized trials covering over 5,000 critically ill patients found that using procalcitonin levels to guide antibiotic discontinuation shortened antibiotic courses by an average of about 1.3 days and was associated with lower mortality overall.¹³ The mortality benefit was modest and did not hold up consistently across subgroups, particularly in patients specifically diagnosed with sepsis or in trials without industry sponsorship. A separate large randomized trial also found that procalcitonin guidance shortened treatment duration and was linked to a decrease in mortality among critically ill patients with presumed bacterial infections.¹⁴

Procalcitonin is not a perfect test. It can rise for reasons other than bacterial infection, and low levels do not always rule out serious infection. But as a guardrail against unnecessarily prolonged courses, particularly in the ICU, it offers one of the few objective signals that it may be safe to stop.

Rapid Diagnostics Are Changing the Timeline

The fundamental tension of empiric therapy stems from a diagnostic gap: traditional blood cultures take one to three days to identify a pathogen, and susceptibility results arrive even later. New technologies are compressing that window.

A randomized trial tested rapid multiplex PCR-based blood culture identification against conventional methods. The rapid test identified organisms in a median of about 1.3 hours after a blood culture flagged positive, compared to over 22 hours with standard techniques. Patients in the rapid-testing group spent fewer hours on broad-spectrum piperacillin-tazobactam and more time on appropriately narrow-spectrum drugs. Treatment of contaminants, a common source of unnecessary antibiotics, dropped from 25% to around 8-11%.¹⁵

Even newer approaches go beyond culture entirely. Metagenomic next-generation sequencing of cell-free DNA in blood can detect pathogens directly from a plasma sample. One study found that this method had a detection rate of about 72% in patients already receiving empiric antibiotics, compared to only about 22% for conventional microbiological tests in the same group.¹⁶ This matters because empiric antibiotics suppress bacterial growth and often make traditional cultures turn up negative, leaving doctors flying blind. A test that works even when antibiotics are already on board could fundamentally change how quickly empiric therapy gets refined.

Machine Learning as a Prescribing Aid

Several research groups are testing whether machine learning models can predict which bacteria a patient is likely to carry and which drugs those bacteria will resist, before any lab results come back. The inputs are the kind of data already collected in most hospitals: age, prior antibiotic use, past culture results, ward location, recent procedures.

A prototype study in Cambodia applied machine learning to predict whether pediatric bloodstream infections would respond to common empiric regimens. The best-performing model predicted susceptibility to ceftriaxone with an area under the curve of 0.80 and susceptibility to ampicillin plus gentamicin with an AUC of 0.74, using only routinely available patient data.¹⁷ Another model focused on urinary tract infections found that recent fluoroquinolone prescriptions were strongly predictive of fluoroquinolone resistance in subsequent cultures.¹⁸

None of these tools have replaced clinical judgment yet. Most are still in the validation phase, and their accuracy depends heavily on local data quality. But the direction is clear: rather than every patient getting the same broad empiric regimen, future prescribing could be personalized based on individual risk factors for resistant organisms.¹⁹

Lessons From COVID-19 Overprescribing

The early months of the COVID-19 pandemic offered a stark illustration of how fear and diagnostic uncertainty drive empiric antibiotic use. At the onset of the pandemic, over 80% of patients hospitalized for COVID-19 received empiric antibiotics, and prescribing for other viral respiratory infections jumped to nearly 70%.²⁰ This happened despite the fact that confirmed bacterial co-infections in COVID patients were relatively uncommon.²¹

Prescribing rates did eventually fall, stabilizing around 35% for COVID hospitalizations by 2022-2023.²² But the early surge illustrated a broader pattern: when doctors face a new or uncertain clinical situation, the default impulse is to cover with antibiotics just in case. That instinct is understandable and sometimes life-saving, but scaled across millions of patients, it accelerates resistance and causes avoidable side effects. The pandemic experience has reinforced the need for better diagnostic tools and clearer stewardship protocols, particularly in the first weeks of any new respiratory threat.

Special Populations and Edge Cases

Empiric therapy looks different depending on who the patient is. In newborns with suspected early-onset sepsis, the standard regimen across neonatal intensive care units in Australia and New Zealand is gentamicin combined with either penicillin or ampicillin, a narrow combination aimed at the organisms most likely to colonize infants during birth. For late-onset sepsis, practices diverge: about 41% of units use empiric vancomycin while 48% prefer flucloxacillin, with the choice depending more on institutional preference than on individual risk factors like gestational age or the presence of a central line.²³

Cancer patients whose white blood cell counts drop dangerously low after chemotherapy (a state called febrile neutropenia) represent another high-stakes scenario where empiric antibiotics are given almost universally. The standard approach involves an anti-pseudomonal beta-lactam, sometimes combined with a glycopeptide like vancomycin. Cochrane reviews have compared different beta-lactams head to head in this population without identifying a clear winner, suggesting the choice often comes down to local resistance data and side-effect profiles rather than any single “best” drug.²⁴

Bacterial meningitis occupies a different niche because the drugs have to cross the blood-brain barrier to reach the infection, which limits the options considerably. Empiric therapy requires agents that are not just active against the likely organisms but can actually achieve effective concentrations in the cerebrospinal fluid.²⁵ This constraint narrows the antibiotic menu and makes the choice of empiric regimen more standardized, typically a third-generation cephalosporin with or without ampicillin and sometimes dexamethasone.

How You Give the Drug Matters Too

Even when you have picked the right antibiotic, how you infuse it can make a difference. Beta-lactam antibiotics kill bacteria most effectively when their concentration stays above a certain threshold for as long as possible. Standard practice has been to give them as a short infusion over 30 minutes, but there is growing evidence that extending the infusion over three or four hours improves outcomes.

A meta-analysis of 24 randomized trials involving over 9,500 critically ill patients found that prolonged infusion of beta-lactams lowered short-term mortality compared to the standard intermittent approach. The evidence was rated high-certainty, which is rare in ICU antibiotic research.²⁶ This is a relatively simple change to implement: same drug, same dose, just delivered more slowly. Many ICUs have already adopted prolonged infusion protocols, but the practice is far from universal.

Outpatient Prescribing and Resistance on the Rise

Empiric antibiotics are not just a hospital issue. In outpatient settings, urinary tract infections are among the most common reasons for empiric prescribing. A patient with symptoms of a UTI is typically given an antibiotic right away, with a urine culture sent to the lab as a backup. The problem is that resistance to the oral antibiotics commonly used for UTIs is increasing, and the implications of prescribing a drug the pathogen can resist are not always clear-cut.²⁷

For uncomplicated UTIs, a mismatched empiric antibiotic sometimes works anyway, because the drug achieves high concentrations in the urinary tract that can overwhelm modest resistance. But for more complicated infections or patients with recurrent UTIs, prior antibiotic exposure, or underlying conditions, a mismatch can mean a return visit, a second course of antibiotics, and sometimes progression to a more serious infection. Checking local antibiograms and factoring in a patient’s own history of resistant organisms become especially important in repeat offenders.