A total of 5602 Salmonella FoodNet records corresponded to isolates with antimicrobial susceptibility results in NARMS. Fifty-three linked records (0.9%) were excluded from further analysis because the submitted specimen did not meet laboratory quality criteria or because the record was a duplicate, leaving 5549 cases in the analysis. Of these, 1105 (20%) showed resistance to ≥1 antimicrobial; resistance was most common to tetracycline (12%), sulfisoxazole (11%), streptomycin (10%), and ampicillin (10%) (Table 1). Two hundred twenty-six isolates (4.1%) had the resistance pattern ACSSuT, followed by ASSuT (n = 157; 2.8%), AAuCx (n = 150; 2.7%), ACSSuTAuCx (n = 32; 0.6%), and CxCip (n = 28; 0.5%). Eighty-five percent of isolates belonged to 1 of 20 common serotypes (Table 2). Serotypes with the highest percentage of resistant isolates were Hadar (73%), I 4,[5],12:i:- (49%), Heidelberg (44%), Typhimurium (34%), and Agona (31%). Heidelberg (16%) and Hadar (11%) were also among the serotypes most associated with invasive infections.

Compared with patients whose isolates showed no resistance, patients whose isolates showed resistance were similar in age, sex, and symptoms (P > .05 for all comparisons) but were less likely to be associated with an outbreak (12% vs 14% P = .03) and more likely to have traveled internationally (16% vs 8%, P < .01) (Table 3). Significant differences occurred among race/ethnicity groups between those with and without resistant infections (P < .01); patients whose isolates showed resistance were more likely to be Black non-Hispanic (143/925 [15%] vs 462/3643 [13%], P = .03) or Asian non-Hispanic (71/925 [8%] vs 144/3643 [4%], P < .01) and less likely to be White non-Hispanic (555/925 [60%] vs 2491/3643 [68%], P < .01). Among travelers, those with resistant infections were more likely to have traveled to Asia (49/149 [33%] vs 32/277 [12%], P < .01) and less likely to have traveled to Mexico (34/149 [23%] vs 94/277 [34%], P = .02). Patients with resistant infections were also more likely to have an isolate source other than stool (189/1105 [17%] vs 503/4444 [11%], P < .01).

Compared with patients whose isolates showed no resistance, patients whose isolates had any resistance were more likely to be hospitalized (31% vs 28%, P = .01), to have a hospitalization length-of-stay ≥3 days (20% vs 16%, P = .01), or to die (1% vs 0.4%, P = .01) on unadjusted bivariate analysis (Table 4). Patients whose isolates had multidrug resistance were also more likely to be hospitalized (34% vs 28%, P < .01) or to have a hospitalization length-of-stay ≥3 days (21% vs 16%, P < .01); similar results were found for patients with isolates resistant to amoxicillin-clavulanic acid, ampicillin, chloramphenicol, streptomycin, sulfisoxazole, tetracycline, all cephems, and resistance patterns including ACSSuT, AAuCx, and ACSSuTAuCx. Patients with isolates resistant to trimethoprim-sulfamethoxazole or ciprofloxacin were not more likely to be hospitalized or to have a length-of-stay ≥3 days, but those whose isolates were resistant to ciprofloxacin were more likely to die (2% vs 0.4%, P < .01). However, patients whose isolates had clinical resistance were not significantly more likely to have any outcome associated with severe disease.

After adjustment for age, sex, race/ethnicity, state, international travel, outbreak association, season, serotype, and specimen source, patients with resistant isolates had higher odds of hospitalization; the association was significant for patients whose isolates showed any resistance (aOR 1.23, 95% CI 1.04–1.46]), multidrug resistance (1.40 aOR, 95% CI 1.12–1.75), or the common resistance patterns ACSSuT (aOR 1.79, 95% CI 1.28–2.50) and AAuCx (aOR 1.73, 95% CI 1.18–2.55) (Figure 1A). Clinical resistance and resistance patterns ASSuT, ACSSuTAuCx, and CxCip were not significantly associated with hospitalization (Figure 1A). Only patterns ACSSuT (aOR 1.70, 95% CI 1.17–2.49) and AAuCx (aOR 1.57, 95% CI 1.02–2.42) were significantly associated with an outcome of hospitalization length-of-stay ≥3 days (Figure 1B).

We found that infections caused by Salmonella with any antimicrobial resistance or multidrug resistance were associated with more severe clinical outcomes, including hospitalization, hospital length-of-stay ≥3 days, and death. This association remained for the outcome of hospitalization after adjustment for demographic factors, international travel, serotype, and isolate source. Moreover, infections caused by Salmonella strains with any resistance or multidrug resistance were also associated with (nonsignificant) aORs above 1.0 for length-of-stay ≥3 days. However, our data analysis suggests that factors other than treatment failure may be responsible for difference in outcomes: infections caused by Salmonella strains with clinical resistance were not significantly associated with hospitalization and had a (nonsignificant) aOR <1.0 for length-of-stay ≥3 days. Furthermore, ciprofloxacin and trimethoprim-sulfamethoxazole were among the only individual antimicrobials that showed no significant association with hospitalization or length-of-stay on unadjusted analysis, although these antimicrobials were recommended for first-line treatment during most of the study period [23].

The lack of association between clinical resistance and poor clinical outcome, along with the known association with overall resistance, indicate that infection with a resistant organism may be a marker for, but not necessarily the direct cause of, a poor clinical outcome. This finding has clinical implications: Clinicians using an antimicrobial that is effective against a resistant Salmonella strain should recognize that such patients are still at increased risk of more severe clinical outcomes than patients with susceptible strains.

There are several possible explanations for an association between resistance and poor clinical outcomes beyond treatment failure. Host factors such as comorbidities or foods consumed may play a role. For example, persons at risk of poorer outcomes may have more exposures that result in infections with resistant strains. Persons with comorbidities may spend more time in hospitals or other healthcare settings, which can be reservoirs of resistant strains [24, 25]. Some serotypes tend to be more resistant than others, and particular serotypes and strains have reservoirs in particular host animals and may more often contaminate foods derived from those animals [26]. Therefore, people’s diets may affect the serotype and resistance profile of their Salmonella strains. Strain factors may also affect outcomes. Virulence varies by serotype [27, 28]. Certain serotypes cause illness disproportionately to their representation in foods, animals, or the environment; some serotypes, such as Dublin or Choleraesuis, have strong associations with both antimicrobial resistance and severe illness [29–31]. Some analyses have suggested a correlation between virulence and resistance [27, 28, 32]. The optimal approach to identifying and defining markers for virulence in Salmonella is still being studied [33].

Our findings are similar to previous US studies that have shown an association between resistance and more severe clinical outcomes. A meta-analysis of 9 US studies found that multidrug resistance was significantly associated with outcomes of bloodstream infection, hospitalization, and death compared with no resistance [14]. The meta-analysis findings are compelling, but many of the individual studies were small or serotype-specific, and they took variable approaches to adjusting for patient and isolate factors. Most patients in these studies had infections that occurred during 1984 to 2013, earlier than many patients in our analysis [5–13]. Therefore, it is noteworthy that the current analysis, which used a robust dataset, more recent data reflecting modern resistance profiles, and a more complete approach to adjustment for other factors, found largely similar results.

This analysis has several limitations. National surveillance data do not include strain type, underlying health conditions, antimicrobial use, duration of illness, or treatment failure. Salmonella isolates were not routinely sequenced until 2015 [15], so we were unable to adjust for the role of specific virulence markers on clinical outcomes. Furthermore, adjustment for serotype is complicated by the number of serotypes capable of causing human illness [34]; our data set contained more than 200 serotypes, but our model only considered the top 10. Therefore, we were unable to directly adjust for less common serotypes known to be associated with resistance, such as Dublin. Furthermore, substantial diversity within serotypes has been appreciated following the implementation of whole genome sequencing (WGS) [35]. Missing data for variables such as travel and race/ethnicity may have influenced the final models. Sensitivity analyses classifying unknown travel status as “no travel” did not affect the statistical significance for any individual antimicrobial or other resistance combination but did alter the directionality (ie, >1.0 vs <1.0) of some aORs with non-significant confidence intervals. Resistance to some antimicrobials was rare overall and typically occurred in combination with resistance to other antimicrobials, making it difficult to isolate the impact of resistance to individual agents. An association between more severe outcomes and clinical resistance is possible and could have been missed; however, our findings suggest that if such an association is present, it is likely not a major contributor to poor outcomes. Finally, we were not able to assess resistance to agents that were not tested pheno-typically. Such resistance is generally believed to be rare based on the assessment of resistance determinants in sequenced Salmonella isolates [36] but could affect the classification of a small number of isolates.

Antimicrobial resistant Salmonella infections in the United States remain associated with worse clinical outcomes. However, clinical resistance alone was not associated with worse outcomes, suggesting that factors other than treatment failure, for example, strain virulence, strain source, or host factors, may be important. These findings underscore the importance of ongoing efforts to reduce and prevent transmission of resistant enteric pathogens, including interventions to limit the entry of resistant strains into the United States and the spread of resistance within agricultural, environmental, and healthcare settings. Further studies are needed to assess the direct and indirect economic costs attributed to resistant Salmonella and to clarify the role of host and strain factors in determining clinical outcomes.