Changes in ABR Over Time

Over a 6-year period between 2017 and 2022, ABR rates were calculated for deduplicated GNB BSI in patients across our local hospitals (n = 10,486). An example of this output, piperacillin/tazobactam resistance in haemato-oncology patients, is shown in Fig. 1.

Fig. 1

Piperacillin/tazobactam resistance in haemato-oncology patients. Top A stacked barplot representing the number of samples in the data; middle a ‘spaghetti plot’ of a random selection of splines fitted to the data that make up the model, and bottom a ‘ribbon’ plot showing the inferred posterior mean (black line) with the 66% and 95% credible intervals (dark and light blue shaded areas) with the quarterly rates shown as dots

The resistance rate increased for meropenem (1.5% to 2.2%, PPI = 88.7%), but reduced slightly (see supplementary material) in all other ABs studied [piperacillin/tazobactam (tazocin), gentamicin, amikacin, cefotaxime, ceftazidime, ciprofloxacin, trimethoprim/sulfamethoxazole (cotrimoxazole) and amoxicillin/clavulanate (coamoxiclav)]. The reductions in resistance rate ranged from 2.3% in ceftazidime (Fig. 2a) (PPD = 94.3%) to 6.9% in trimethoprim/sulfamethoxazole (Fig. 2b) (PPD = 98.5%). This larger reduction in resistance may be related to the removal of trimethoprim as first-line treatment for lower urinary tract infection in local community guidelines from April 2017, which resulted in a reduction in trimethoprim resistance in E. coli from community urine samples [11].

Fig. 2

All deduplicated GNB BSI isolates (n = 10,486): a percent resistance to ceftazidime reduced from 12.9% (95% CI 10.9–15.3) in 2017 to 10.6% (95% CI 8.8–12.9) in 2022. b % Resistance to trimethoprim/sulfamethoxazole reduced from 32.9% (95% CI 28.3–36.4) in 2017 to 26.0% (95% CI 22.7–29.4) in 2022

Informing Empiric AB Choice in Specific Patient Groups

When examining resistance to specific ABs in specific patient populations, the 95% CI tends to widen due to the reduced sample size. For example, piperacillin/tazobactam resistance in GNB BSIs in ICU patients has a much wider 95% CI compared to the whole hospital data, as the former is based on 695 positive blood cultures compared to 10,486 for the latter (Fig. 3a, b). Over time, piperacillin/tazobactam resistance rates have increased for ICU patients and haemato-oncology patients (Fig. 3b, c), while reducing over the whole hospital population (Fig. 3a). Clearly, therefore, using the mean resistance rate for piperacillin/tazobactam over the 6-year period to inform empiric AB choice would underestimate the current resistance rate in the ICU and haemato-oncology populations, while using only the last year or few months data would be excessively influenced by random month-to-month variation in small patient groups. Our model overcomes this issue as the time series data allow us to estimate current ABR rates, and the uncertainty, while the model borrows information from the earlier data.

Fig. 3

Piperacillin/tazobactam resistance for all patients: top compared to ICU patients (middle) and haemato-oncology patients (bottom)

Following this example, the posterior means between piperacillin/tazobactam resistance rates in GNB BSI isolates from the ICU population and the whole hospital population are different (27.3% vs. 13.4%, respectively), and the 95% CIs do not overlap, suggesting that this difference is reliable. The probability that ICU patients have higher resistance rates can be computed simply as the proportion of curves that are higher in models of the ICU population than the whole hospital population at this timepoint, in this case, PPI = 99.9%.

In contrast, the difference between the posterior mean piperacillin/tazobactam resistance rates for BSI GNBs from ICU and haemato-oncology patients is small (27.3% vs. 30.7%) and there is a considerable overlap in the 95% CIs. This is reflected in the calculated posterior probability of 67.5% that piperacillin/tazobactam resistance rate is lower in the ICU population compared with the haemato-oncology population (e.g. PPD = 67.5%) or conversely there is a 32.5% probability that the resistance rate is greater in ICU patients than in haemato-oncology patients (e.g. PPI = 32.5%).

The situation for ceftazidime is similar, with a higher (and increasing) rate of resistance in both ICU and haemato-oncology patients compared to the whole hospital population. Both ICU and haemato-oncology patients have a higher proportion of potentially AmpC hyper-producing “SPACE” (Serratia, Pseudomonas, Acinetobacter, Citrobacter and Enterobacter) isolates compared to the whole population (ICU 38% vs. 25%, PPD = 99.3%) and haemato-oncology (30% vs. 25%, PPD = 87.0%) which explains a proportion of the higher resistance rates.

The increase in ceftazidime resistance over time is likely greater in haemato-oncology patients (15.2% to 24.5%, PPI = 89.4%) than ICU (17.6% to 20%, PPI = 66.3%), and is due to both increased resistance in non-E. coli isolates include in SPACE organisms, (Fig. 4) and in an increase in the proportion of non-E. coli isolates (E. coli reduced from 44% to 35%, PPD = 82%, in haemato-oncology, while in ICU the proportion of E. coli remained at 27%).

Fig. 4

Percent resistance to ceftazidime in all isolates (left column) and non-E coli isolates (right column) located in either (top row) ICU (bottom row) or haematolo-oncology

Calculating an Escalation Antibiogram (EA) for Specific Patient Groups

To calculate an EA, our Bayesian model has been applied for any given second-choice AB to all isolates resistant to the first-choice AB. Table 1 reports these data for the whole hospital population expressed as a most likely rate of resistance with 95% credible intervals. In this population, if moving away from piperacillin/tazobactam, we would predict 27% (95% CI 18–24) if moving to gentamicin and 31% (95% CI 21–40) if switching to ciprofloxacin.

Table 1 Posterior mean resistance estimates with 95% credible intervals for the whole hospital population; tables for haemato-oncology, paediatrics, GI or urinary source infections are in the supplementary material

We can also calculate the PPInf between pairs of alternative antibiotics. This is shown in Fig. 5 using data for ICU patients assuming piperacillin/tazobactam resistance. When making a clinical decision on antibiotic selection, the probability of inferiority is only useful when both alternatives have a suitably low rate of resistance.

Fig. 5

For all patients (top) and ICU patients (bottom) with assumed resistance to piperacillin/tazobactam. The estimated resistance rate to 6 antibiotic options with the 66% and 95% credible interval is shown on the left. The centre and right plots each show a head to head comparison between one antibiotic (amikacin or ciprofloxacin) and the other five antibiotics. This gives the estimated difference in probability of resistance, for example, in ICU patients there is 40% probability of higher resistance in amikacin compared to ciprofloxacin (right), only marginally favouring amikacin, while there is an 83% probability of higher resistance in co-trimoxazole compared to ciprofloxacin (right), favouring the use of ciprofloxacin

It is worth noting that GNB BSI resistance rates to meropenem are low in this region (~ 2%), where meropenem is invariably superior as an empiric choice. Our analysis has therefore focused on other ABs as meropenem-sparing alternatives. Similarly within our region, resistance rates to newer antibiotics such as ceftazidime-avibactam or meropenem-vaborbactam are currently too low to allow inclusion in our analysis.

Escalation Antibiogram Example 1

In severe Gram-negative infections, two of the most commonly used β-lactams are piperacillin/tazobactam (first-line AB for neutropenic sepsis and widely used in ICU) and ceftazidime (a useful alternative to piperacillin/tazobactam especially in non-severe penicillin allergy). In our region, piperacillin/tazobactam and ceftazidime resistance rates among GNB BSIs are similar: 13.4% (95% CI 10.8–16.1) and 10.6% (95% CI 8.9–12.7), respectively. The merits of the addition of an aminoglycoside to a β-lactam has been much debated [12, 13], but will inevitably depend on local resistance and co-resistance rates. We will look at the effect of resistance to piperacillin/tazobactam and ceftazidime on the probability of resistance to gentamicin (our mostly widely used aminoglycoside) and amikacin (which is rarely used locally) in two specific patient groups, ICU patients and haemato-oncology, and compare these results with those generated from the whole hospital population.

When comparing a wide group of second line options when adding to, or switching from, piperacillin/tazobactam or ceftazidime (Tables 2 and 3), ABs are ordered in preference for whole hospital populations, showing the percent resistant with 95% CI, and with the PPInf to that in the column on its left in the table. Values under 50% indicate the AB is superior to that on its left.

Table 2 Escalation antibiogram for piperacillin/tazobactam (pip/taz) resistant isolates showing resistance rate and the posterior probability of inferiority (PPInf) to the antibiotic choice to the immediate left (e.g., Gentamicin v. Amikacin, Ciprofloxacin v. Gentamicin, etc.) across the whole hospital and in subgroups from ICU and Haematology/OncologyTable 3 Escalation antibiogram for ceftazidime-resistant isolates showing resistance rate and the PPInf to the antibiotic choice to the immediate left (e.g. gentamicin vs. amikacin, ciprofloxacin vs. gentamicin, etc.) across the whole hospital and in subgroups from ICU and haemato-oncology

Resistance rates to both aminoglycosides are much higher in isolates resistant to either ceftazidime or piperacillin/tazobactam, with resistance rates vary from 14 to 39% depending on the patient group and antibiotic combinations, compared to from 3% to 13% aminoglycoside resistance in all the isolates.

Within the whole hospital population, amikacin remains the best option for patients switching from either ceftazidime or piperacillin/tazobactam, with a 97% and 95% probability of superiority compared to gentamicin, although there is a greater relative increase in amikacin resistance compared to gentamicin (see Tables 2 and 3).

Within the ceftazidime and piperacillin/tazobactam resistant isolates, both the ICU and haemato-oncology populations differ from the whole hospital population, with lower rates of aminoglycoside resistance in ICU and higher rates of resistance in haemato-oncology patients (Fig. 6). In ICU patients, amikacin is likely to be superior, PPSup = 69.8% if switching from piperacillin/tazobactam and PPSup = 71% if switching from ceftazidime. For the haemato-oncology patients, there is little difference between amikacin and gentamicin, (PPSup = 52%), although there is a noticeably higher rate of resistance for both when switching from ceftazidime compared to piperacillin/tazobactam (also see plots in supplementary material).

Fig. 6

Percent resistance to gentamicin assuming piperacillin/tazobactam resistance in (top) all isolates 27.0% (95% CI 18.5–34.5), (middle) ICU 19.2% (95% CI 7.1–37.0), (bottom) haemato-oncology 33.5% (95% CI 14.6–57.5)

A difference between the whole population and ICU or haemato-oncology populations are seen in a range of other antibiotics. Within the ICU patients, ciprofloxacin, trimethoprim/sulfamethoxazole and gentamicin have a similar probability of sensitivity, which is noticeably lower than in the whole hospital population. While, within the haemato-oncology cohort, resistance rates to amikacin and gentamicin are higher than in the general population, and resistance rates to ciprofloxacin and trimethoprim/sulphamethoxazole are similar, despite both being used in some prophylaxis regimes. The resistance rates for all options are high (30–40%), so, in a neutropenic patient, meropenem would be a more suitable alternative.

It is clear, therefore, that sub-population analysis will be required by those intending to apply the EA in clinical practice.

Escalation Antibiogram Example 2

Having noted how changes in resistance over time differ between species, for example cefotaxime resistance in E. coli and non-E. coli (see supplementary material), we confirmed that this effect was also present in piperacillin/tazobactam-resistant isolates (Fig. 7). We then looked at two clinical groups with a high proportion of E. coli infections (patients over 80 years old and patients with a urinary source of infection) to determine if they differed from the whole hospital population (Tables 4 and 5).

Fig. 7

Top Cefotaxime resistance in all isolates assuming piperacillin/tazobactam resistance was stable over time 43.7% (95% CI 33 to 53) PPD = 46.1%. Bottom Cefotaxime resistance in E. coli assuming piperacillin/tazobactam resistance reduced from 29.2% (95% CI 20.4–38.6) to 18.3% (95% CI 11–28) PPD = 95.5%

Table 4 Escalation antibiogram for piperacillin/tazobactam-resistant isolates showing resistance rate and the PPInf to the antibiotic choice to the immediate left (e.g. gentamicin vs. amikacin, ciprofloxacin vs. gentamicin, etc.)Table 5 Escalation antibiogram for ceftazidime resistant isolates showing resistance rate and the PPInf to the antibiotic choice to the immediate left (e.g. gentamicin vs. amikacin, ciprofloxacin vs. gentamicin, etc.)

Overall, the proportion of BSI caused by E. coli has very likely decreased from 59.1% (95% CI 54.9–63.8) at the start of 2017 to 51.4% (95% CI 47.1–55.5, PPD = 99.6%) at the end of 2022. Within patients over 80 years old, and patients with a BSI of urine or renal tract source we have seen a similar decrease in the proportion of E. coli BSI, to a current rate of 59.7% (95% CI 54.3–64.4) from 68% (95% CI 63.5–72.9, PPD = 99.3%) and 57.8 (95% CI 48.5–64.4) from 65.6% (95% CI 58.2–71.9, PPD = 94.5%) respectively.

When switching from piperacillin/tazobactam (Table 4), it appears that amikacin is the best option, but, in circumstances where an aminoglycoside could not be used (e.g. poor renal function), cefotaxime is comparable to ciprofloxacin and superior to co-trimoxazole for urinary source infections.

We would not usually think of empirically changing from piperacillin/tazobactam to cefotaxime, as the former is usually considered to be broader spectrum, and it is often assumed that resistance to piperacillin/tazobactam would also confer resistance to 3rd generation cephalosporins. While AmpC enzymes can confer resistance to both classes, when ESBL-producing isolates are reported as piperacillin/tazobactam-resistant, this can be due to a range of β-lactamases which are not routinely identified in clinical isolates. β-lactamases, including OXA-1, inhibitor-resistant TEM, or the high levels of TEM-1, can result in piperacillin/tazobactam resistance, independently of 3rd generation cephalosporins resistance [14].

Assuming ceftazidime resistance (Table 5) showed that amikacin is clearly superior in the urinary source group, whereas, for the over-80s, amikacin or gentamicin are broadly comparable. Due to the high resistance rates of co-trimoxazole and ciprofloxacin, meropenem would be suitable if an aminoglycoside could not be used.