Introduction

Patients admitted to intensive care units (ICUs) present challenging and complex clinical problems. The estimated risk for serious infection is 5–10 times greater than for patients on general medical wards owing to three major factors: severe underlying disease, including multiple illnesses, malnutrition, extremes of age and immunosuppression; invasive medical devices, such as endotracheal tubes for mechanical ventilation and intravascular and urinary catheters, which provide entry portals for pathogens; and crowding, especially in neonatal ICUs, with consequent proximity to other colonised or infected patients, increasing the risk of cross-infection [1–3].

Antimicrobial resistance is a critical variable of ICU outcomes, co-determining patient morbidity, mortality and cost, at least in the major teaching centres where this topic has been largely investigated [4–12]. Kollef, in the USA, found an infection-related mortality rate of 17.7 % among 486 patients receiving appropriate empirical antimicrobial therapy versus 42 % among 169 receiving inappropriate antimicrobial treatment (p < 0.001) [13]. The single major reason for antibiotic therapy being classed as “inappropriate” was the presence of bacteria that had inherent or acquired resistance to the regimen. Others have found similar associations, particularly in bloodstream infections and sepsis [14–20], with mortality shown to increase progressively for each hour’s delay in initiating adequate therapy after the onset of hypotension [21]. In the few countervailing studies, where an association between antibiotic resistance and mortality was not confirmed, few patients received microbiologically inappropriate therapy, owing to early recognition of resistance and/or timely adjustment of the regimen(s) [22, 23]. Beyond its impact on mortality, initial inappropriate antibiotic therapy is also associated with an extended length of stay for ICU patients [24–26].

It is less clear whether these relationships, demonstrated in teaching centres with a complex patient mix, hold true for the smaller ICUs of district general hospitals or in the context of different countries’ cultures of prescribing and microbiological testing. We therefore present here the results of a multicentre, retrospective, observational study covering five ICUs in the Veneto region of north-east Italy, four of them in small hospitals and the fifth in a regional centre. The study had three main goals: first, to test whether, as elsewhere, there was a relationship between treatment inadequacy and clinical outcomes; second, to examine the adequacy of the first-line antimicrobial therapy prescribed and the principal reasons for any inadequacy; and, third, to verify the appropriateness of microbiological testing performed in the participating ICUs.

Patients and methods

Study location and patients

The study was conducted from 2002 to 2010 at five general hospitals in the Veneto region of north-east Italy. Four hospitals were in small towns within 50 km of Vicenza and one in Vicenza itself, which is located between Venice and Verona. When this study was performed, the Vicenza hospital (hospital 5) ICU had 14 beds, admitted approximately 700 patients per year, and was in a 1,050-bed regional hospital; a further 5-bed high-dependency provision for post-surgical care was excluded. Hospital 1 (165 beds) admitted approximately 350 patients per year to its six-bed medical–surgical ICU; hospital 2 (400 beds) had a ten-bed general ICU admitting c. 450 patients per year; hospital 3 (350 beds) had a six-bed general ICU admitting c. 240 patients per year; hospital 4 (220 beds) had a seven-bed ICU admitting c. 340 patients per year. There was a total of 43 ICU beds, accounting for 10.8 % of ICU provision in the Veneto and for 1.5 % of 3,739 Italy’s total intensive care bed provision, as of 2005 [27].

Data collection

Patients admitted into the participating ICUs from 15 May 2002 to 10 June 2010 were assessed. Data input was performed manually in Microsoft Office Excel, with the following information recorded: hospital record number; gender; date of birth; date of hospital admission; date of ICU admission (if different); age at ICU admission, and main diagnosis at admission. Any other diagnoses indicated in the clinical records and constituting: a co-morbidity, a chronic disease directly related to ICU admission, or a secondary pathological event that occurred during the ICU stay was also recorded. For statistical analysis, diagnoses were classified into main categories, all as recognised in the WHO International Statistical Classification of Diseases and Related Health Problems [28]. The date of the primary outcome (death, or transfer to another unit) was recorded. Additionally, for patients transferred from the ICU to other units in the same hospital, the dates of transfer were recorded until the final outcome (death or discharge to home). The duration of ICU stay and entire hospital stay were calculated separately. For each antibiotic course, the regimen and dates of initiation and cessation were recorded. An antibiotic treatment was defined as empirical when it was initiated on the basis of a clinical suspicion of infection and when the causative microorganism and its antibiotic susceptibility were not yet known. Fungal infections were excluded. A sample was considered clinically relevant when it had been taken from a body site related to the reported infection.

Inadequate antimicrobial treatment was defined (based upon, for example, Kollef [29] and Cosgrove and Carmeli 30]) as the microbiological documentation of infection that was not being adequately treated at the time when the causative micro-organism and its antibiotic susceptibility became known, and included:

  1. 1.

    The absence of any agent directed against the family or genus of the micro-organism present

  2. 2.

    The administration of an antimicrobial agent to which the particular isolate was resistant

  3. 3.

    The complete lack of antimicrobial treatment

  4. 4.

    The lack of adherence to minimum requirements in antibiotic administration (i.e., proper dosing, monitoring of drug levels when appropriate, and avoidance of unwanted drug interactions)

A regimen was defined as adequate if it adequately covered all pathogens present in a sample taken at the time of clinical diagnosis (T0). Adequacy was considered to be non-assessable if there was no T0 sample, if no pathogen was grown from a T0 sample, or if there was no concordance between the type of specimen sent to the laboratory and the patient’s clinical presentation (e.g. clinically diagnosed septic shock in post-surgical patients, but where the first isolates were grown from surgical wound swabs taken many days after the initiation of empirical antibiotic treatment; or cases of sepsis where the only microbiological examinations performed were on bronchial aspirates). Cases where only questionable pathogens (principally coagulase-negative staphylococci) were isolated were reviewed individually and discounted unless therapy was escalated on the basis of the microbiological result, implying that the organism was thought to be clinically significant.

Statistical analyses

Normally or near-normally distributed variables were reported as means with standard deviations and were compared using Student’s t test or analysis of variance with the Bonferroni correction for multiple comparisons. Non-normally distributed continuous data were reported as medians with interquartile ranges (IQRs) and were compared using the Mann–Whitney U test or the Kruskal–Wallis test. Spearman’s rho correlation coefficient was calculated to measure the association at the ordinal level between mortality rates and their associated rates of inadequacy of first-line antimicrobial therapy. Kaplan–Meier methods were used to estimate survival rates during follow-up, whereas the log-rank test was used to test the equality of survivor functions. Exploratory univariate analysis for several variables was performed to identify possible predictors of hospital mortality. Multivariate logistic regression analysis was conducted to investigate independent predictors of hospital mortality. Results of logistic regression analysis are reported as adjusted odds ratios (AORs) with 95 % confidence intervals (CI). All statistical analyses were performed using STATA 10.1 (StataCorp LP, College Station, TX, USA) and a two-sided p < 0.05 was routinely considered to be significant.

Ethics

The study was performed in accordance with the recommendations guiding physicians in biomedical research involving human subjects adopted by the 18th World Medical Association, Helsinki, Finland, 1964, and later revisions [31]. As it was performed retrospectively on specimens that were collected as part of the routine sampling required for the microbiological assessment of patients admitted into ICUs, there was no possible risk to any of the patients reviewed, nor any possible modification of their treatment. Consequently, individual consent was not needed. The Institutional Sanitary Board of each hospital approved the protocol and confirmed that submission to their ethics committee was not required, provided that the principal investigator (PB) was personally responsible for the security of patient-identifiable data.

Results

Patients

The study reviewed 911 patients admitted into the five ICUs (Table 1): 570 (62.6 %) were men and 341 (37.4 %) women. Eighty-eight (9.6 %) had diabetes mellitus, 45 (5.0 %) chronic renal failure, and 35 (3.9 %) cirrhosis of the liver. Five hundred and fifty-two (60.6 %) were admitted to an ICU with a medical diagnosis, 206 (22.6 %) with a surgical diagnosis, and 153 (16.8 %) following major trauma. Their mean age upon ICU admission was 65.8 ± 16.2 years (range, 14–93); those admitted to the hospital 5 ICU were significantly younger (p < 0.001) than those admitted elsewhere, partly reflecting a larger proportion of trauma patients. The median duration of ICU stay was 17.0 days (IQR, 8.0–29.0), with an inter-hospitals difference approaching significance (p = 0.079), whereas the median total length of hospital stay was 25.0 days (IQR, 14.0–44), with significant inter-centre variation (p < 0.001).

Table 1 Characteristics of the study patients
Full size table

Antibiotic treatment

A total of 3,549 antimicrobial treatments were prescribed in the 5 ICUs over the study period. Of these, 3,470 (97.8 %) were parenteral and 79 (2.2 %) oral. Seven hundred and fifty-seven patients (83.1 %) received empiric antibiotic courses (1,223 courses in total, 34.5 % of all antimicrobial treatments; Table 2). Monotherapy was used in 30.2 % of empirical courses, with combination therapy used in 69.8 %. The commonest empirical combinations were piperacillin/tazobactam plus a glycopeptide or linezolid (52 patients, 13.4 %), a carbapenem plus a glycopeptide or linezolid (47 patients, 12.1 %), a cephalosporin plus a glycopeptide or linezolid (27 patients, 7.0 %), piperacillin/tazobactam plus a fluoroquinolone (25 patients, 6.4 %), a cephalosporin plus metronidazole (23 patients, 6.0 %), and piperacillin/tazobactam plus metronidazole (21 patients, 5.4 %).

Table 2 Initial empiric antibiotic therapy received by the patients surveyed
Full size table

Combination therapies included two antibiotics in 329 cases (85.0 %), three in 55 cases (14.2 %), and four in 3 cases (0.8 %). Cephalosporins (148 courses) accounted for 40.1 % of all empirical monotherapies, with cefazolin (first-generation) in 66 (44.6 %), cefotetan (second-generation) in 4 (2.7 %), cefotaxime, ceftazidime, and ceftriaxone (third-generation) in 69 (46.6 %), and cefepime (fourth-generation) in 9 (6.1 %). Other frequently prescribed monotherapies were piperacillin/tazobactam (74 courses, 20 % of all monotherapies) and other penicillin/β-lactamase inhibitor combinations (68 courses, 18.4 %).

Only 126 of the 388 combination regimens (32.5 %) comprised a broad-spectrum Gram-negative β-lactam plus an anti-MRSA agent (a glycopeptide, usually teicoplanin, or linezolid); 91 (12.2 %) of the first-line empirical regimens were irrational or redundant poly-pharmacy, commonly comprising a combination of a β-lactam with anti-anaerobe activity (i.e. a β-lactamase inhibitor combination or a carbapenem) with metronidazole.

The median duration of the first-line empirical therapy was 11.0 days (IQR, 6.7–19.0) for patients with bacteraemia, 9.0 days (5.0–14.0) for medical patients and 10.0 days (7.0–17.0) for surgical patients. Although there is a growing trend towards shortening treatment durations, particularly in Northern Europe, these longer courses are typical of Italy in the study period and are not out of line with many international guidelines [32].

Laboratory data

There was often a poor match between the site of infection indicated in the patient record and the specimens from which organisms, if any, were grown by the laboratory. Moreover, there were frequently long delays between the clinical diagnosis and any result becoming available to the treating clinician(s).

At the four smaller sites (hospitals 1–4), respiratory samples accounted for >50 % of all specimens with a pathogen grown, and for 74 % and 82 % at hospitals 2 and 3, respectively (Fig. 1). Blood and (especially) urine were rarely sampled, even when an infection was believed to involve these sites. Thus, at hospitals 2 and 3, urine accounted for only 5.6 % and 8.7 % of total microbiological investigations, respectively. These patterns seem to reflect a practice of performing surveillance cultures of respiratory secretions and basing therapy upon these, rather than of undertaking microbiological investigations of actual infections.

Fig. 1
figure 1

Sites of clinically diagnosed infections (grey) compared with the number of specimens sampled (black), by site. Numbers indicate hospitals

Full size image

Clinical specimens yielding an organism were collected at the initiation of empirical antibiotics (T0) from only 251 of the 911 patients (27.6 %). Sixteen of these 251 did not receive antibiotics, as they were considered to be colonised rather than infected (n = 13) or died early (n = 3), leaving 235 patients who had a T0 specimen and an assessable empirical antibiotic treatment. This total reduced to 212 after the exclusion of 23 patients, whose T0 sample was from a body site different to the infection recorded in the patient’s notes. Samples yielding reported organisms were taken within 10 days of therapy initiation from a further 361 patients (39.6 %), whereas in the remaining 299 cases (32.8 %), the interval between the initiation of antibiotic therapy and the first sample with a reported organism was >10 days, or there was no relationship between the type of specimen from which any organism was grown and the patient’s clinical setting (Table 3). The median interval between the initiation of empirical antibiotic therapy and the availability of a first (post-infection) antibiotic sensitivity result was 7 days (IQR, 3.0–14.0), with significant variation amongst the five sites (p < 0.001). The lag between the arrival of a growth-yielding sample at the laboratory and the availability of the result varied between sites from 3 to 4.5 days, meaning that around half of this overall 7-day delay was between the clinical diagnosis of infection and the specimen being sent to the laboratory for microbiology. It follows that many of the patients were already receiving antibiotics at the time the first culture was taken, potentially compromising pathogen recovery and meaning that many were nearing the end of their antibiotic course when any microbiological results became available.

Table 3 Timing of initial clinical specimens
Full size table

The lack of a T0 organism may be because no specimen was sent to the laboratory, or because no organism was grown by the laboratory. Discriminating between these scenarios in the hospital record systems proved difficult, but, for a random sample of 23 patients lacking culture results, we could identify 6 who had a relevant-site T0 sample from which the laboratory failed to grow a pathogen, 8 who had only a T1–T10 specimen failing to yield growth, and 9 who had no evidence of any specimen being sent to the microbiology laboratory within 10 days of clinical diagnosis.

In total, 313 isolates from clinical samples (regardless of site and apparent relevance) were collected at T0 from the 235 patients starting empirical treatment. In 147 cases (62.6 %) the organism(s) proved susceptible to the antibiotic regimen initiated, whereas 88 (37.4 %) patients had bacteria resistant to the regimen initiated. The lowest proportion of resistance was at hospital 1 (25.0 %) and the highest at hospital 2 (42.4 %). Resistance to the empirical therapy was more prevalent (192 out of 347, 53 %, p < 0.001) amongst patients who had initial isolates collected during the T1–T10 period, again with the lowest proportion (49.1 %) at hospital 4 and the highest (61.5 %) at hospital 2. The differential in resistance, between patients with a T0 vs a T1–T10 initial sample was moderately significant, even in the bacteraemia subset, where 35 out of 54 (64.8 %) isolates from patients with a T1–T10 sample were resistant to the empirical therapy given vs 14 out of 34 (41.2 %; p = 0.098) isolates from T0 samples, whereas the difference in resistance between the bacteraemic vs non-bacteraemic patients in the whole series was not significant (p = 0.1891; Table 4). Among the 212 patients who had clinically relevant T0 samples, resistance to the empirical therapy given was observed in 74 (34.9 %), with the lowest proportion (22.6 %) at hospital 1 and the highest (40.6 %) at hospital 2 (p = 0.2277). A greater proportion of resistance to empirical therapy (142 out of 266, 53.4 %), was seen in cases with an initial T1–10 sample, with the lowest rate (43.4 %) at hospital 4 and the highest (70.0 %), again, at hospital 2 (p = 0.01). Among bacteraemic cases, 6 out of 24 T0 isolates (25.0 %) were resistant compared with 13 out of 25 (52 %) among those collected from T1–10. There was little obvious demographic difference between the groups, with a first relevant-site sample at T0, T1–10 and T > 10 (or no relevant sample at all), with average ages of 66.3, 65.8, and 66.5 years and medical:surgical:trauma ratios of 73.5:15.5:11.0; 59.0:19.9:21.1; 71.2:14.8:14.0 respectively.

Table 4 Bacterial resistance (R) or susceptibility (S) to empirical therapy administered
Full size table

The frequent lateness of the microbiological data may explain the small proportion of cases (282 out of 757, 37.2 %) in whom empirical regimens were adjusted based upon susceptibility results. The vast majority of these changes (252 out of 282, 89.4 %) were escalations, meaning the addition of further agents or switches to broader-spectrum agents; first-line empirical antibiotic was stepped down in only 30 cases (10.6 %).

Outcomes

Two hundred and twenty-seven patients (24.9 %) died during their ICU stay and 316 (34.7 %) during their entire hospitalisation. One hundred and forty-three of the ICU deaths (63 % of all ICU deaths) could reasonably be related to infection. Patient primary outcome data in relation to treatment adequacy for the 212 cases with a relevant-site T0 clinical specimen is displayed in Table 5. Among the 74 (34.9 %) whose empirical antibiotic(s) failed to cover the organisms subsequently identified there were 36 ICU deaths (48.6 %) vs 26 deaths (18.8 %) among the 138 receiving therapy that covered all pathogens present (p < 0.001). This pattern was maintained among patients whose first relevant specimen was taken in the T1–T10 interval, where there was 43 % mortality among those receiving inadequate therapy vs 23 % among those receiving adequate antimicrobial therapy (OR = 1.84; 95 % CI, 1.3 to 2.5). In this case, however, it is impossible to distinguish whether inadequacy was against the initial pathogen, against its resistant progeny, or against a secondary invader. Overall mortality rates among patients with a first relevant-site sample at T0, T1–T10 and T > 10 (or no relevant sample at all) were 29.2 %, 19.9 % and 27.6 % respectively.

Table 5 Patient primary outcomea in relation to treatment adequacy among patients with a baseline relevant specimen
Full size table

The adequacy of the initial regimen did not significantly affect the duration of ICU stay (p = 0.93; Fig. 2), partly because survivors who were hospitalised for extended periods were balanced by patients who died early.

Fig. 2
figure 2

Duration of intensive care unit stay in relation to the adequacy of empirical treatment

Full size image

Development of septic shock was a significant predictor of mortality as was the patient’s age (p < 0.001). Non-survivors were also more likely to have had acute renal failure upon admission (p < 0.001). By contrast, those admitted because of traumatic shock were more likely to survive (p < 0.001), perhaps owing to a higher probability of receiving adequate antibiotic treatment, given to 36.6 % of trauma patients vs 30.9 % of other patient categories.

The commonest pathogens isolated from bloodstream samples and their associated rates of inadequate antimicrobial treatment were P. aeruginosa (n = 23; 80 % inadequacy), MRSA (n = 19; 80 % inadequacy) and E. coli (n = 14; 77 % inadequacy). The large number of MRSA is unsurprising: European Antimicrobial Resistance Surveillance Network (EARS-net) data (http://www.ecdc.europa.eu) show that the MRSA rate among bloodstream S. aureus fluctuated between 33.2 and 39.4 % throughout the study period.

Despite extensive cephalosporin use (as mentioned above) only 2 cases of Clostridium difficile diarrhoea were recorded, although it should be cautioned that diarrhoeal patients were not routinely screened for this pathogen during the study period, leading to likely under-recording.

Discussion

Studies of antibiotic inadequacy and its consequences in patients who are severely ill have largely been undertaken in teaching centres [13–21, 33–35], particularly in the USA [13, 15, 17, 33]. We investigated whether their general conclusion—that inadequate empirical therapy is associated with increased mortality—also applied to smaller centres in the Veneto region.

Assessing treatment adequacy proved challenging, owing to the many patients in whom microbiology was carried out improperly or belatedly. Clinical specimens were collected at initiation of empirical antibiotics (T0) for only 31.0 % of patients (235 out of 757) starting an initial empirical antibiotic course, and only 28 % (212) had a sample taken from the reported infection site. In rather more cases (347, 45.8 %), initial samples were taken between T1 and T10, whereas for 299 cases (39.5 %), the interval to the first sample was >10 days, or there was no concordance between any laboratory specimen and the patient’s clinical setting. Resistance to the initial antibiotic therapy was significantly more prevalent amongst T1–T10 isolates than among those collected at T0, regardless of whether comparison was irrespective of body site (55.3 vs 37.4 %) or solely from isolates from the relevant site (53.4 vs 34.9 %). Similar patterns—with greater resistance in T1–T10 vs T0 samples—were seen for the subset of bacteraemia patients (64.1 % vs 41.1 % for isolates from any body site and 52.0 vs 25.0 % amongst bloodstream isolates). Greater resistance rates among “late” isolates may reflect the selection of resistance in the original pathogen, or super-infection by more resistant organisms during therapy.

When only the 212 empirically treated patients with a relevant T0 sample were analysed (Table 5), 138 treatments (65.1 %) were assessed as being adequate, with 18.8 % deaths in the ICU, vs 74 (34.9 %) being judged inadequate, with 48.6 % deaths; p = <0.001). Total mortality among these 212 patients with timely microbiological investigation was 29.2 % compared with 27.6 % among the patients who had very belated microbiological investigation (>10 days) or no investigation at all. These two groups were well matched in terms of average age and proportions of medical vs surgical vs trauma cases; the overall similarity in outcome may well reflect the fact that, even where microbiological investigation was performed, therapy was rarely changed. Mortality was lower (19.9 %) among the patients who had a first relevant sample in the T0–T10 period, but this group contained a higher proportion of trauma patients, who tended to have better outcomes anyway.

This study, covering five small ICUs in the Veneto region, thus confirms a significant association between inadequate empirical antimicrobials and ICU mortality. S. aureus and resistant Gram-negative bacteria were the pathogens most frequently associated with poor outcomes, also as seen elsewhere [14, 15, 17, 21, 33–35]. Furthermore, and in keeping with a recent meta-analysis [36], ICU infections following trauma had lower mortality, perhaps because most trauma patients are younger and have fewer co-morbidities. In contrast to several published studies [8, 14, 18, 20, 24–26], we found no relationship between treatment adequacy and duration of ICU stay, although this calculation may be confounded by how duration is counted for patients who die early.

The frequent lack of prompt microbiological investigation is the core finding of this study. In many cases, clinicians’ decision seemed to depend upon surveillance sampling for bacterial colonisation of the lower airways rather than direct microbiological investigation of the clinically diagnosed infections, often with an excess of antibiotics. Such overtreatment seems widespread in Italy [37, 38] and elsewhere [39, 40].

Even when relevant specimens were collected, they were often collected late, meaning that the organisms grown may have been secondary colonists, and that susceptibility results only became available around the time when primary empirical treatment was ending, or even afterwards. This may explain the infrequency of treatment de-escalation based on laboratory results. Even hospital 5—where laboratory results were available earlier—in all but 2 cases, changes to initial empirical antibiotic treatment were escalations, not de-escalations.

In summary, despite its limitations (e.g. being retrospective, exclusion of fungal infections and the difficulty of evaluating empirical therapy in patients whose microbiological investigation was inadequate), this study provided a clear picture of sub-optimal microbiological testing and antibiotic use in the five ICUs. There was frequent antibiotic misuse, inappropriate empirical treatment, and high variability in (generally overlong) treatment duration and a considerable need for the ICUs to improve specimen-taking and the use of the microbiology laboratory. Notably, the ICUs lacked local antibiotic practice guidelines, which represent one tool for clinicians to manage patient and stewardship needs. Clinicians should be aware that any transient clinical benefit achieved by overtreatment is counterbalanced by collateral damage and detriment to the community as a whole via increased selection pressure for resistance. These issues are becoming even more serious and urgent with the recent and extensive dissemination in Italy of Klebsiella pneumoniae with KPC carbapenemases [20, 41, 42]. Poorly directed antibiotic use may have helped to drive this dissemination, which saw the proportion of carbapenem-resistant bloodstream K. pneumoniae in Italy rise from 1–2 % in 2006–2009 to over 30 % in 2013–2014 (http://www.ecdc.europa.eu). Most of these isolates are, however, clonal [43] and it therefore seems likely that infection control failures are a greater issue, a view supported by the observation that near-identical strains of carbapenemase-producing K. pneumoniae were reduced in prevalence in Israel by improved infection control rather than stewardship changes. Carbapenemase-producing K. pneumoniae remain rare at the ICUs included here (5–8 isolates, hospital-wide, per annum from 2010–2012 in hospital 5, rising to 14 in 2013, 18 in 2014 and 34 in the first half of 2015), but are hugely more prevalent in a major teaching centre just 50 km away (data not shown).