We conducted comprehensive BSI surveillance in two rural provinces: Sa Kaeo (SK) in eastern Thailand bordering Cambodia and Nakhon Phanom (NP) in northeast Thailand bordering Laos People’s Democratic Republic. Both provinces are agrarian with estimated populations of approximately 526,000 and 734,000, respectively, from National Economic and Social Development Board (NESDB) projections based on 2010 census data.^20,21 This study included data between 2008 and 2014 from all 20 hospitals of the two provinces (16 district, two military [10–140 beds], and two provincial hospitals [225–327 beds]), and included all laboratories performing hemoculture.²²

As clinically indicated, hemocultures were performed using two different conditions, one optimized for aerobic growth and the other for enhanced growth of mycobacteria and other fastidious organisms as previously reported. A modification was applied in October 2011; the mycobacterial type bottles were not routinely processed but available on physician request.²³ Hemoculture bottles inoculated at district or military hospitals were kept at 15–30°C and transported to the provincial laboratory within 24 hours before placement in the BacT/ALERT 3D system (BioMérieux, Marcy-l’Étoile, France). Alarm-positive bottles were subcultured on chocolate, sheep blood, and MacConkey agars and incubated overnight at 35°C. Standard biochemical tests were performed for organism identification.²⁴ From 2012, Analytical Profile Index Microbial Identification Kits (BioMérieux) were used when standard testing was not definitive. Escherichia coli- and K. pneumoniae-positive cultures that grew a likely contaminant in the same bottle were excluded from our case counts. Repeat positive cultures that grew the same species within 30 days were excluded.

Antibiotic susceptibility testing was performed using a modified Kirby–Bauer disk diffusion method.²⁵ To improve screen test sensitivity, ESBL screening used two single-disc diffusion tests: ceftazidime (CAZ) and cefotaxime (CTX) with zones of inhibition of ≤ 22 or ≤ 27 mm, respectively, were considered positive.²⁵ Confirmatory testing used a combination disc method. Ceftazidime/Clavulanic acid (CAZ/CLA) or CTX/CLA discs were placed on the same culture plates as the single discs. A ≥ 5-mm increase in combination disc zone diameter compared with single disc alone confirmed ESBL production. Confirmation testing was routinely performed in SK. Nakhon Phanom started confirmatory testing in 2014. We defined an ESBL-producing pathogen as an isolate that was either confirmation or screen test positive when confirmatory testing was not available.

Multiplex polymerase chain reaction for five common genetic mechanisms of resistance (bla_NDM-1, bla_OXA, bla_IMP, bla_KPC, and bla_VIM) were performed as previously described.²⁶

Likely contaminants that grew in the same hemoculture bottle (51 E. coli- and 20 K. pneumoniae-positive cultures) were excluded from case counts. These were Staphylococcus spp. other than Staphylococcus aureus (n = 28), Gram-positive bacillus (n = 16), Bacillus spp. other than Bacillus anthracis (n = 15), Streptococcus viridans group (n = 8), and other (n = 4).

Escherichia coli and K. pneumoniae cases were classified as hospital onset (HO) when the hemoculture was drawn after day 3 of hospital stay, with day of admission being day 1.²⁷ Community onset (CO) cases were described by a positive hemoculture collected on or before day 3 of hospital stay. Preculture antibiotic use was any antibiotic received within 72 hours of culture determined by patient report and review of medical records. Recurrent infections were defined as positive hemoculture for the same pathogen more than 30 days after the initial positive culture. Mortality outcome was based on information available at the time of hospital discharge and, therefore, reflects solely in-hospital mortality.

The CO incidence rate of E. coli and K. pneumoniae bacteremia was calculated using age-specific population estimates from the NESDB.²⁰ For the period 2007–2009, the 2010 NESDB age distribution was applied to the 2007–2009 NESDB overall provincial population estimates,²⁰ as official intercensal estimates were not available.²¹ To calculate HO incidence cases, we used the number of hospitalizations from 2008 to 2014 from hospital registration databases (S. Naorat and B. Piralam, personal communication). In SK, four district hospitals were missing 2008 and 2009 data. Together, these hospitals account for ≤ 15% of annual hospitalizations in our surveillance system, so we imputed hospitalization data from 2010 data for these four hospitals. Military hospitals were also excluded from HO incidence calculations, as hospitalization data were not available. Military hospitals account for < 1% of hospitalizations in each year of surveillance.

The chi-squared or Fisher exact tests were used as appropriate. Ninety-five percent CIs on incidence estimates were calculated from the normal approximation to the Poisson distribution.

The CDC Human Subjects Review Office determined that this protocol constituted routine public health activities and did not involve human subject research (CGH 2014-273).

Between 2008 and 2014, 97,832 hemocultures were performed, with 12,813 (13.1%) positive for a pathogen. This excludes 3,859 cultures that grew a likely contaminant and pathogen in the same culture bottle. A total of 3,338 E. coli infections (1,595 from SK and 1,743 from NP) and 1,086 K. pneumoniae infections (529 from SK and 557 from NP) were identified. Annual E. coli bacteremia case counts increased from 373 in 2008 to 593 in 2014, with corresponding increases in annual incidence from 32.9 (2008) to 51.6 (2014) per 100,000 persons. Likewise, K. pneumoniae case counts increased from 116 in 2008 to 178 in 2014, with corresponding increases in incidence from 10.2 (2008) to 15.5 (2014) per 100,000 persons.

Extended-spectrum β-lactamase screening was performed for 98.7% (3,293/3,338) of E. coli and 97.5% (1,059/1,086) of K. pneumoniae isolates (Figure 1). Comparisons of single-disc screening versus combination disc confirmatory tests demonstrated that screen tests had a high sensitivity (98.4% in E. coli and 95.7% in K. pneumoniae), specificity (93.8% in E. coli and 92.1% in K. pneumoniae), and overall agreement (95.5% [1,324/1,387] for E. coli and 93.0% [427/459] for K. pneumoniae). We, therefore, included screen test only positive isolates with confirmatory test positives as “ESBL producing.” Overall, 26.8% (883/3,293) of E. coli and 27.4% (290/1,059) of K. pneumoniae isolates were ESBL producing (Figure 1).

Extended-spectrum β-lactamase–producing capability among E. coli bacteremia cases ranged from 23.8% to 31.9% by age group; however, ESBL-producing K. pneumoniae was significantly more common among children aged 0–4 years compared with older persons (P < 0.01, Table 1). Prevalence of ESBL in E. coli increased over time: from 19% to 22% in 2008–2010 to approximately 30% from 2011 to 2014 (P-value for trend = 0.02). Prevalence of ESBL among K. pneumoniae cases had no apparent trend over time (P = 0.09). There were no statistically significant differences between the two provinces for the percentage of E. coli and K. pneumoniae that were ESBL producing (Table 1).

Persons aged ≥ 50 years accounted for 77.3% (2,547/3,293) of E. coli bacteremia cases and 78.1% (690/883) of ESBL-producing E. coli cases. Similarly, 73.2% (775/1,059) of K. pneumoniae bacteremia cases and 66.6% (193/290) of ESBL-producing K. pneumoniae cases were in persons aged ≥ 50 years. Among children aged < 5 years, neonates ≤ 28 days accounted for 50.0% (36/72) of E. coli bacteremia, 60.9% (14/23) of ESBL-producing E. coli, 58.7% (27/46) of K. pneumoniae cases, and 63.0% (17/27) ESBL-producing K. pneumoniae.

The majority of E. coli (93.6%) and K. pneumoniae cases (87.6%) were CO; however, HO cases were twice as likely to be ESBL-producing compared with CO cases. For E. coli, 51.7% (106/205) of HO cases were identified as ESBL versus 25.2% (777/3,083) of the CO cases (P < 0.01). For K. pneumoniae, 55.9% (71/127) of HO cases were ESBL versus 23.6% (219/928) of CO cases (P < 0.01) (Table 1).

Hospitalization in the past year was significantly more frequent among patients with ESBL-producing E. coli and K. pneumoniae (63.7% and 71.6%, respectively) than in patients with non-ESBL–producing organisms (36.3% and 28.4%, respectively). The percentage of ESBL E. coli did not differ between provincial and district level hospitals (27.1% versus 25.7%, P = 0.37), but ESBL K. pneumoniae was significantly more common in provincial level hospitals than district hospitals (30.6% versus 22.6%, P = 0.01).

Among those with available outcome data, the mortality rate was twice as high among case patients with ESBL-producing E. coli (14.5%, 56/387) versus 7.1% (85/1,202) for non-ESBL–producing E. coli (P < 0.001), and this difference persisted after age stratification (P < 0.01). In fact, mortality was nearly three times higher among patients aged 70 years and older with ESBL-producing E. coli (20.0%, 26/130) than among patients with non-ESBL–producing E. coli (7.4%, 33/446). Among children aged 0–19 years, there were no deaths among 35 patients with non-ESBL E. coli compared with two deaths among 13 patients (15.4%) with ESBL-producing E. coli. But for K. pneumoniae case patients, the mortality rate was not different between those with ESBL-producing and non-ESBL–producing infections (19.1% [25/131] and 14.4% [56/390], respectively) (P = 0.197).

Overall CO ESBL-producing E. coli bacteremia incidence during 2008–2014 was 9.8 cases per 100,000 persons per year (95% CI: 9.1–10.5). This incidence increased from 5.4 (95% CI: 4.1–6.7) in 2008 to 12.8 per 100,000 persons per year (95% CI: 10.7–14.9) in 2014. The community onset ESBL-producing K. pneumoniae bacteremia incidence from 2008 to 2014 was 2.7 per 100,000 (95% CI: 2.4–3.1), from a low of 2.2 (95% CI: 1.3–3.0) in 2008 to a high of 4.0 (95% CI: 2.8–5.1) in 2012.

The estimated incidence of CO ESBL-producing E. coli was highest among persons aged ≥ 70 years, and this age group showed the most pronounced increase from 34 cases/100,000 population in 2008 (95% CI: 19–48) to 79 cases/100,000 population (95% CI: 59–100) in 2014 (Figure 2). Incidence among persons aged 50–69 years also increased over time from 9/100,000 (95% CI: 5–13) in 2008 to 26–30/100,000 (95% CI: 20–37) in 2012–2014. Incidence was lower among persons aged 20–49 and 0–19 years and remained stable over time. The overall estimate for these age groups during the 2008–2014 period was 4.2/100,000 (95% CI: 3.5–4.9) and 0.7/100,000 (95% CI: 0.4–1.0), respectively. This pattern is consistent with the increase in overall E. coli case counts during this period which was primarily limited to patients older than 50 years (Figure 2). Similarly, CO ESBL-producing K. pneumoniae incidence was highest in persons aged ≥ 70 years (12.9 cases/100,000 [95% CI: 9.6–16.3]), followed by persons aged 50–69 years (5.6 cases/100,000 [95% CI: 4.4–6.7]) and persons aged 20–49 years at 1.4 cases/100,000 (95% CI: 0.2–2.3), and lowest in persons aged 0–19 years at 0.6 cases/100,000 (95% CI: 0.3–0.9) (Figure 2). However, unlike E. coli, CO ESBL-producing K. pneumoniae incidence did not increase significantly in any of these age categories over time (Figure 2). The number of cases of ESBL-producing K. pneumoniae remained stable despite an increase in total K. pneumoniae case counts over time. Higher K. pneumoniae case counts in 2013–2014 were offset by a lower percentage of ESBL-producing infections (25%) compared with that in 2011–2012 (31–32%).

The incidence of HO ESBL-producing E. coli during the 7-year period varied between 0.13 and 0.30 cases/1,000 hospitalizations in SK. This was significantly higher than that in NP, where the incidence was 0.06 cases/1,000 hospitalizations in 2008, which dropped to 0.01 in 2009 and increased to 0.12 cases per 1,000 hospitalizations in 2014 (P < 0.01 for the trend between 2009 and 2014). Initially, the incidence of HO ESBL-producing K. pneumoniae from 2008 to 2014 was higher in SK at 0.12 (95% CI: 0.09–0.16) cases/1,000 population compared with 0.05 cases/1,000 population (95% CI: 0.03–0.07) in NP (P = 0.03). However, HO K. pneumoniae bacteremia incidence increased significantly over time in NP (P = 0.03) and by 2014 estimates from the two sites were similar; in 2014, HO ESBL-producing K. pneumoniae was 0.11 (95% CI: 0.0–0.2) in NP and 0.13 (95% CI: 0.0–0.2) in SK.

Most E. coli (88.8%; 2,500/2,815) and K. pneumoniae (86.0%; 772/898) case patients did not receive antibiotics in the 72 hours before blood draw; however, pre-culture antibiotic use was three to four times more common among ESBL-producing case patients than in non-ESBL–producing case patients (Table 2). The most common pre-culture antibiotics used were cephalosporins, which was significantly higher among ESBL-producing cases (Table 2).

There were significantly different antibiotic susceptibility profiles between ESBL and non-ESBL CO E. coli infections. Extended-spectrum β-lactamase CO E. coli infections had significantly lower susceptibility to amoxicillin (43.9% versus 79.2%), ceftazidime (27.9% versus 98.9%), cefotaxime (11.2% versus 98.3%), gentamycin (50.1% versus 88.5%), ciprofloxacin (27.3% versus 81.8%), and trimethoprim/sulfamethoxazole (26.6% versus 43.2%) (Figure 3). Similarly for K. pneumoniae CO infections, there was lower susceptibility for amoxicillin (46.2% versus 92.7%), ceftazidime (36.4% versus 99.1%), cefotaxime (28.0% versus 98.4%), gentamycin (54.6% versus 98.5%), ciprofloxacin (44.4% versus 92.9%), and trimethoprim/sulfamethoxazole (37.0% versus 87.0%) (Figure 3).

Among HO cases, a significant difference in susceptibility was seen between ESBL and non-ESBL–producing E. coli for third-generation cephalosporins (4.8% versus 94.9% for cefotaxime and 12.5% versus 95.9% for ceftazidime), fluoroquinolone (32.4% versus 78.4% in ciprofloxacin), and aminopenicillins (34.6% versus 77.6% in amoxicillin) (all P < 0.001) (but not for aminoglycosides, carbapenems, and antibiotic combinations involving a β-lactamase inhibitor). More than 85% of HO ESBL-producing E. coli and K. pneumoniae were susceptible to piperacillin/tazobactam, amikacin, and carbapenems. Similar differences in susceptibility to cephalosporins, fluoroquinolones, and aminopenicillins were seen among HO K. pneumoniae cases.

Of 864 ESBL-producing E. coli, two were carbapenem-resistant Enterobacteriaceae (CRE) (one in 2012 encoding genes other than five tested for, and one in 2014 carrying bla_NDM-1 gene). Of 277 ESBL-producing K. pneumoniae, three were CRE (two cases with bla_NDM-1 in 2013 and 2014, and one case in 2012 encoding genes not tested for). Of 2,400 non-ESBL–producing E. coli, two were CRE (one case in 2011 encoding bla_IMP and another in 2014 encoding genes not covered). There were no non-ESBL–producing K. pneumoniae CRE cases.

Recurrent bacteremia was noted in 3.8% (126/3,274) of E. coli case patients. These 126 patients had a total of 271 episodes; 14.3% (18/126) had ≥ 3 episodes of E. coli bacteremia. Of the 271 bacteremia cases, 105 (38.7%) were ESBL, which was similar to the overall percent of ESBL among E. coli infections. At least a quarter (25.9% or 14/54) of the recurrent bacteremia was new, as a non-ESBL infection was preceded by an ESBL infection (Figure 4). Recurrent bacteremia was present in 2.7% (28/1,043) of patients with K. pneumoniae and 28.6% (8/28) were ESBL-producing, similar to the overall percentage among K. pneumoniae.