Increased Incidence of Antimicrobial-Resistant Nontyphoidal Salmonella Infections, United States, 2004–2016

Salmonella is a major cause of foodborne illness in the United States, and antimicrobial-resistant strains pose a serious threat to public health. We used Bayesian hierarchical models of culture-confirmed infections during 2004–2016 from 2 Centers for Disease Control and Prevention surveillance systems to estimate changes in the national incidence of resistant nontyphoidal Salmonella infections. Extrapolating to the United States population and accounting for unreported infections, we estimated a 40% increase in the annual incidence of infections with clinically important resistance (resistance to ampicillin or ceftriaxone or nonsusceptibility to ciprofloxacin) during 2015–2016 (≈222,000 infections) compared with 2004–2008 (≈159,000 infections). Changes in the incidence of resistance varied by serotype. Serotypes I 4,[5],12:i:- and Enteritidis were responsible for two thirds of the increased incidence of clinically important resistance during 2015–2016. Ciprofloxacin-nonsusceptible infections accounted for more than half of the increase. These estimates can help in setting targets and priorities for prevention.

Salmonella is a major cause of foodborne illness in the United States, and antimicrobial-resistant strains pose a serious threat to public health. We used Bayesian hierarchical models of culture-confi rmed infections during 2004-2016 from 2 Centers for Disease Control and Prevention surveillance systems to estimate changes in the national incidence of resistant nontyphoidal Salmonella infections. Extrapolating to the United States population and accounting for unreported infections, we estimated a 40% increase in the annual incidence of infections with clinically important resistance (resistance to ampicillin or ceftriaxone or nonsusceptibility to ciprofl oxacin) during 2015-2016 (≈222,000 infections) compared with 2004-2008 (≈159,000 infections). Changes in the incidence of resistance varied by serotype. Serotypes I 4, [5],12:i:-and Enteritidis were responsible for two thirds of the increased incidence of clinically important resistance during 2015-2016. Ciprofl oxacin-nonsusceptible infections accounted for more than half of the increase. These estimates can help in setting targets and priorities for prevention.
Salmonella infections in the United States, and whose only known reservoir are humans (2,16,17,23). In this article, we use the term Salmonella to refer to nontyphoidal Salmonella.

National Antimicrobial Resistance Monitoring System
The National Antimicrobial Resistance Monitoring System (NARMS) is a collaboration among CDC, the US Food and Drug Administration, the US Department of Agriculture, and state and local health departments to monitor resistance among enteric bacteria isolated from humans, retail meat, and food animals (16,24). Public health laboratories in 50 state and 4 local health departments submit every 20th Salmonella isolate received from clinical laboratories to CDC for antimicrobial drug susceptibility testing (16,19,24).

Resistance Categories
We defined clinically important resistance as resistance to ceftriaxone, nonsusceptibility to ciprofloxacin, or resistance to ampicillin on the basis of the following criteria: third-generation cephalosporins (e.g., ceftriaxone) and fluoroquinolones (e.g., ciprofloxacin) are used for empiric treatment of severe infections; fluoroquinolones are not recommended for children; and ampicillin is useful for susceptible infections (2). We defined and ranked from highest to lowest 3 mutually exclusive categories of clinically important resistance (Appendix Figure 1, https:// wwwnc.cdc.gov/EID/article/27/6/20-4486-App1. pdf) (20): ceftriaxone/ampicillin resistance (because all ceftriaxone-resistant isolates are ampicillin-resistant); ciprofloxacin nonsusceptibility (nonsusceptible to ciprofloxacin but susceptible to ceftriaxone); and ampicillin-only resistance (ampicillin-resistant but susceptible to ceftriaxone and ciprofloxacin). We included ciprofloxacin-nonsusceptible isolates that were ceftriaxone-resistant in the ceftriaxone/ampicillin category because they are of greatest public health and treatment concern. Many isolates in each category had resistance to other agents. We defined multidrug resistance as resistance to >3 classes of antimicrobial agents (16,19).

Bayesian Hierarchical Model to Estimate Changes
We used 2004-2016 data from LEDS, NARMS, and the US Census Bureau as input in the models (16,17,25). For LEDS, we used the number of culture-confirmed infections by state and year (state-year). We combined serotyped isolates other than Enteritidis, Typhimurium, Newport, I 4, [5],12:i:-, and Heidelberg into an "other" category. We assigned unserotyped and partially serotyped isolates from each state into the 6 serotype categories (Enteritidis, Typhimurium, Newport, I 4, [5],12:i:-, Heidelberg, and other) on the basis of the average proportion of serotyped isolates in each category from 2004-2016. For NARMS, we used resistance proportions among fully serotyped isolates per state-year. We used US Census population data for each state-year to express incidence per 100,000 persons per year (25).
A similar Bayesian hierarchical model approach was used from a previous study to estimate the incidence of resistant infections (20,26). However, we found a Poisson model for LEDS data better captured the uncertainty of Salmonella incidence observed at the state-year level instead of the normal distribution used in our previous study (20). The model incorporated the random effects of state, year, and state-year interaction to borrow strength from contiguous states and previous years (20,(26)(27)(28). Alaska and Hawaii were excluded because they are not adjacent to any state; the District of Columbia was also excluded, which began submitting isolates to NARMS in 2008 (16,19,20). We used an approach similar to our previous study to make adjustments for data from Florida, which reported low numbers of isolates compared with its 6 closest states (17,18,20).
We applied the models to generate estimates (referred to as posterior estimates) for Salmonella infection incidence rates, resistance proportions, and resistant infection incidence rates (referred to as resistance incidence) by state-year for each of the 6 serotype categories by using Markov chain Monte Carlo simulations (20,(26)(27)(28). For each serotype category, we estimated resistance incidence for overall clinically important resistance, the 3 mutually exclusive categories of clinically important resistance, and multidrug resistance. For all Salmonella, we calculated overall estimates by summing estimates across the 6 serotype categories. We calculated state-year resistance incidence estimates per 100,000 persons per year as estimated incidence for state-year × estimated resistance proportion for state-year. For estimation of resistance incidence by geographic region, we used the 4 US Census region categories (Midwest, Northeast, South, and West) and aggregated posterior estimates of resistance incidence by year for all states in each region (25). For each resistance category, we calculated mean estimates and 95% credible intervals (CrIs) from posterior estimates and mean crude rates by year for the 48 states and those stratified by serotype and region categories ( Figure 1; Appendix  for an overall sideby-side comparison (20,25,26).
To assess changes in resistance incidence, we compared the mean resistance incidence from 2015-2016 with that from two 5-year reference periods during 2004-2016: 2004-2008 and 2010-2014. These reference periods are consistent with those used in NARMS annual reports to assess changes in resistance percentages (16). All 50 states have participated in NARMS since 2003; the 2004-2008 period is the early years of nationwide participation and the 2010-2014 period is the recent past. For each resistance and serotype category, we calculated the difference between the posterior estimates of resistance incidence for 2015-2016 and those for each year in the 5-year reference periods for each region to obtain the mean difference and the 95% CrIs. We did not assume homogeneous Figure 1. Estimated annual incidence of culture-confirmed nontyphoidal Salmonella infections with any clinically important resistance, by serotype and region, United States, 2004-2016. Estimated changes in resistance incidence (mean and 95% credible intervals of the posterior differences per 100,000 persons per year) were derived using Bayesian hierarchical models. Crude resistance incidence rates were derived by multiplying infection incidence and resistance proportion for state-year. Any clinically important resistance was defined as resistant to ceftriaxone, resistant to ampicillin, or ciprofloxacin nonsusceptible. The "other" category comprised serotypes other than Enteritidis, Typhimurium, Newport, I 4, [5],12:i:-, and Heidelberg. US Census regions were used to define 4 geographic regions. NTS, all nontyphoidal Salmonella serotypes. rates across multiple years using this approach. For all Salmonella, we calculated the change in resistance incidence, which represents the net change (increase or decrease), for each resistance category, by summing the estimated changes derived for the 6 serotype categories. We describe statistically significant changes (i.e., in which the 95% CrIs do not include 0).

Extrapolating to the US Population
We multiplied the mean estimates of culture-confirmed infections by 29, which is the estimated number of total infections for every culture-confirmed infection in the general population, to estimate the total number of resistant infections for each period and changes in total resistant per 100,000 persons per year during 2015-2016 compared with the reference periods for each resistance category (1). We used the average 2015-2016 population estimates for the 50 states (322 million) to extrapolate to the US population (25).

Incidence by Year and Region, 2004-2016
For each resistance category, the trend lines were smoother with model-derived annual estimates of resistance incidence compared with crude rates, particularly when stratified by serotype and region (Figure 1; Appendix . Crude rates tended to be lower than model-derived estimates because many state-year resistance proportions used in calculating crude rates were 0 because of small sample sizes, whereas the model tended to pull estimates away from 0. Overall, most crude rates were within modelderived 95% CrIs.

Changes in Resistance Incidence, 2015-2016 versus 2010-2014
The mean annual incidence of Salmonella infections with clinically important resistance did not change compared with the previous 5 years. However, the mean annual incidence of ciprofloxacin-nonsusceptible Salmonella infections increased by 0.29 (95% CrI 0.02-0.52)/100,000 persons (Table 2); by region, the increase was significant only in the Midwest (Figure 3). Ciprofloxacin-nonsusceptible Enteritidis infections increased by 0.16 (95% CrI 0.02-0.32)/100,000 persons, accounting for 57% of the increase in ciprofloxacin-nonsusceptible infections (Appendix Table 3).

Extrapolation to the US population
Compared with the number of infections for 2004-2008, an estimated ≈63,000 more Salmonella infections with clinically important resistance occurred each year during 2015-2016, from an average of ≈159,000 to ≈222,000; more than half were ciprofloxacin-nonsusceptible (Table 3). Compared with the number of infections for the previous 5 years, an estimated ≈56,000 more Salmonella infections with clinically important resistance occurred each year during 2015-2016; more than half were ciprofloxacin-nonsusceptible.

Discussion
Our analysis indicates that the incidence of resistant The increased incidence of ciprofloxacin-nonsusceptible Salmonella infections during 2015-2016 compared with incidence for both 2004-2008 and 2010-2014 is a concerning trend. Serotype Enteritidis contributed the most to this increase. Although the incidence of infections with Enteritidis, the most common serotype, has not changed significantly in >10 years, the percentage of ciprofloxacin-nonsusceptible infections has increased almost steadily (11,16). Chicken and eggs have been the main domestic sources of Enteritidis infections (29,30). About 20% of Enteritidis infections are linked to international travel, which is an 1668 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 27, No. 6, June 2021 important source of ciprofloxacin-nonsusceptible Enteritidis infections (8,31,32). The incidence of infections with clinically important resistance and ciprofloxacin-nonsusceptibility caused by serotypes categorized as other was higher during 2015-2016 than during 2004-2008. Some of these serotypes are emerging or have concerning levels of resistance, including Dublin, Infantis, Kentucky, Hadar, and Agona (16,24,33). Some have been associated with resistance, invasive illness, or both (11,19,23,33).
The decrease in resistant Typhimurium infections might be related to the simultaneous increase in I 4, [5],12:i-infections, which some call monophasic Typhimurium (16,18,34). In the 1990s, MDR Typhimurium infections increased markedly in Europe and then in the United States (35,36). Most isolates from these infections that underwent phage typing were definitive type 104 (14,21,35,36). Isolations of this strain have decreased globally; the reasons are not known (36) Changes in resistance incidence by resistance category and serotype varied by geographic region, with significant increases in most regions for serotypes I 4, [5],12:i:-and Enteritidis. An increase in the incidence of I 4, [5],12:i:-infections with multidrug and ampicillin-only resistance occurred in all 4 regions, with highest increase in the West and Midwest. Pork products have been associated with I 4, [5],12:i:-infections with resistance to ampicillin, sulfonamide, streptomycin, and tetracycline in the West (34,37). The regional pattern of pork consumption has reflected the regional pattern of pork production, which is highest in the Midwest; 8 of the 10 states with the highest production of swine are in the Midwest (38,39). A study showed that MDR I 4, [5],12:i:-strains from swine in the Midwest during 2014-2016 were typically resistant to ampicillin, sulfonamide, streptomycin, and tetracycline and probably part of a European clade that has spread in the United States and elsewhere; these strains harbored plasmid-mediated resistance genes, which can be transmitted horizontally to other bacteria (34). This trend could partly explain the widespread increase in the incidence of MDR I 4, [5],12:i:-infections. International travel could have contributed to an increase in the incidence of ciprofloxacin-nonsusceptible Enteritidis infections, which increased in 3 regions and was highest in the Northeast. International travel has increased since 2014, and residents of northeastern states accounted for more than one third of US travelers during 2015-2016 (40). In the United Kingdom, an increase in these infections has been linked to international travel and imported foods (41). In the United States, ciprofloxacin-nonsusceptible strains of Enteritidis and other serotypes have been isolated from imported seafood (42). Plasmid-mediated quinolone-resistance genes have been detected among ciprofloxacin-nonsusceptible isolates in the United States; these genes might contribute to spread of fluoroquinolone nonsusceptibility (43).
Our use of a Bayesian hierarchical model improved the estimates, as shown by the smoothing of resistance incidence and temporal change lines, by addressing issues related to missing and sparse stateyear data (20,26). Our method of calculating the aver- age difference in incidence between groups of years is more refined than approaches using a negative binomial model because it does not assume homogeneous resistance incidence rates across multiple years (11,44). It is therefore less likely to underestimate the variability of estimated changes. However, this analysis is subject to the same limitations described in previous reports, including unmeasured sources of bias and uncertainty derived by combining data from separate unlinked surveillance systems (20,26).
Our estimates of significant changes were limited to comparisons with the reference periods used to assess changes in resistance percentages in NARMS annual reports (16). Our choice to compare a recent 2-year period with earlier 5-year periods balanced the need to assess the most current situation with the need for sufficient data to assess significant changes. Because of the low percentage of isolates showing resistance to trimethoprim/sulfamethoxazole (<3%) or decreased susceptibility to azithromycin (<1%), an important agent used to treat serious infections, we did not provide estimates for these agents (2,16,18,20). We included infections resistant to ceftriaxone and nonsusceptible to ciprofloxacin in the ceftriaxone/ampicillin-resistance category; they represented only 0.3% of Salmonella isolates submitted to NARMS. The fact that some ciprofloxacin nonsusceptible infections were not included in the ciprofloxacin nonsusceptible category further supports our finding that ciprofloxacin-nonsusceptible infections increased during the study period. Increasing use of culture-independent diagnostic tests by clinical laboratories can change the submission of isolates to public health laboratories and reporting of infections (11); these changes warrant adjustments in future analyses (20).
We multiplied estimates of culture-confirmed infections by 29 to account for undiagnosed infections. However, resistant infections are associated with more severe illness, so they might be more likely to be detected (3)(4)(5)(6). Thus, the appropriate multiplier (the ratio of total infections to cultureconfirmed infections) for resistant infections might be <29. To calculate undiagnosed Salmonella infections, multipliers of 12 for persons <5 years of age and 23 for persons > 65 years of age have been reported (45). Although children <5 years of age have the highest incidence of Salmonella infections, older adults might disproportionately account for resistant infections because they are more likely to have serious illness and be hospitalized (4,5,(44)(45)(46)(47); therefore, a multiplier of 23 might be an appropriate choice. However, we chose 29 because it was used in a previous estimate of the total number of Salmonella infections in the population (1) and because persons 5-64 years of age account for most cultureconfirmed infections reported to CDC and most isolates with clinically important resistance submitted to NARMS (4,18,44,45). We did not attach uncertainties to the extrapolated total number of resistant infections and changes in that number because uncertainties of the multiplier are not known. Although resistance incidence can vary by demographic subgroup, geographic region, time, and other factors, we did not include additional uncertainties from the extrapolation to the US population using the average 2015-2016 population estimates for the 50 states (19,21,22,46,47).
Estimates of changes in resistance incidence can help identify trends of greatest concern to set priorities for prevention. Analyses that include the varying distributions of infections by demographic subgroups, season, and recent travel could inform serotype-specific, regional, and source-targeted prevention strategies (5,11,21,22,31,(44)(45)(46)(47)(48). The increasing use of whole-genome sequencing by public health laboratories to characterize Salmonella strains will enhance surveillance of antimicrobialresistant Salmonella from human and nonhuman sources (49). Antimicrobial agents contribute to resistance wherever they are used, including in food animals and humans (50). A One Health approach can help in detecting and controlling antimicrobial resistance, which is a complex and multifaceted problem that affects humans, animals, and the environment (50).