The CDC’s EIP program is a partnership between the CDC and 10 state and local health departments and academic partners to conduct comprehensive, active, population-based surveillance for a variety of infections and conditions, including candidemia [17-19]. The Tennessee EIP program began surveillance for candidemia in 2011 in 9 East Tennessee counties and expanded to 17 East Tennessee counties in 2017, for a total population under surveillance of 1.5 million persons. All laboratories that serve the population in the catchment area are recruited to report blood cultures positive for Candida spp to the health department. Periodic audits of laboratory records are conducted to ensure completeness of reporting.

A case of candidemia was defined as a blood culture yielding Candida spp in a surveillance area resident. Repeat positive cultures within 30 days were considered part of the same case. Surveillance officers reviewed medical records for all cases and used a standardized case report form to collect demographics and basic clinical information. Information on history of IDU was collected starting in 2014.

For this analysis, we categorized candidemia cases that occurred between January 1, 2014 and September 30, 2018 as either IDU-associated or non-IDU based on a surveillance officer’s review of the medical chart. An IDU-associated case was defined as a candidemia case with documentation of active IDU of any substance in the 12 months before candidemia diagnosis in the medical record. Documentation of IDU and the substances injected was typically found in the patient’s history, infectious disease consultation history, or social worker consultations. Simply having a discharge diagnosis (International Classification of Diseases, Tenth Revision code) documenting IDU was insufficient evidence to classify cases, because the time at which the patient last used injection drugs was not documented. A non-IDU case was defined as a candidemia case without such documentation.

In addition to the routinely completed candidemia case report form, we used a secondary form to collect additional information on all IDU-associated cases for which medical records were available (n = 77), as well as on a subset of non-IDU cases (n = 103) ≥12 years of age that were frequency-matched to IDU-associated cases by hospital of diagnosis and selected using a random number generator. This secondary form documented which substances patients were injecting, illicit drugs used by any route, the misuse of prescription drugs (whether or not they were obtained from an illicit source), drug panel results, and treatment history. Substance use and route data were collected from patient histories and physician notes in the physical exam, infectious disease consultation, or social worker consultation. Laboratory results and treatment records provided the remaining data. Drugs prescribed to the patient for medical purposes were not documented unless the patient histories specifically mentioned the inappropriate use of those drugs. The secondary form also documented a more detailed medical history, including hospitalization history, reasons for hospitalization, and sites of infection other than bloodstream. Only the first case of candidemia in a given person was included in this analysis.

Candidemia cases were classified into 1 of 3 patient groups. Cases with positive blood cultures obtained on or after 3 days of hospital admission were categorized as hospital-onset. Healthcare-acquired, community-onset (HACO) cases were those with positive blood cultures obtained <3 days after admission, in which patients had healthcare exposure in the 90 days before candidemia diagnosis. Community-onset cases were those with positive blood cultures obtained <3 days after hospital admission but lacking recent healthcare exposure. Recent healthcare exposures were defined as an emergency room visit or hospitalization in the 90 days before the candidemia diagnosis, any surgery in the 90 days before diagnosis, or receipt of a central line in the 2 days before diagnosis.

To ascertain 30- and 90-day mortality, we matched all study patients with Tennessee death records. We reviewed medical records to determine whether the person died during their candidemia hospitalization or postdischarge.

χ² tests were used to compare proportions, and a Kruskal-Wallis test was used for the comparison of medians. Significance was defined as P < .05. All data were analyzed using SAS 9.4 (SAS Institute, Cary, NC).

United States Census Bureau data were used to calculate overall population-based rates of candidemia [20]. The number of PWID in the surveillance area was estimated from the Substance Abuse and Mental Health Services Administration (SAMHSA) National Survey on Drug Use and Health (NSDUH) from 2016 to 2017 [21]. The SAHMSA estimates for number of people who ever reported injecting drugs are available at the state level. We estimated the number of PWID in the counties involved in candidemia surveillance using county-level data collected by TDH on conditions associated with IDU in 2017, including hepatitis C cases (acute or chronic), opioid-related fatal overdoses, and opioid-related nonfatal overdoses (including data from hospitalizations, emergency department visits, ambulatory surgeries, and certain diagnostic services). We used an average of the proportion of these 3 conditions that occurred in the catchment area to estimate the proportion of all Tennessee PWID who reside in the surveillance area. The NSDUH data estimates the number of people who have ever reported injecting drugs; however, data on the number of people that injected in the past 12 months was not available. The National Health and Nutrition Examination Study (NHANES) also collects self-reported data on IDU. Based on a previous study using NHANES data, in the South region (as designated by the US Census Bureau) 6.7%–31.6% of respondents who reported having ever injected drugs had done so in the preceding 12 months [22]. We used this range to estimate the proportion of PWID in the catchment area who had injected drugs within the past year.

During January 1, 2014–September 30, 2018, we identified 637 candidemia cases in 599 people; 82 had a history of IDU in the 12 months before candidemia diagnosis. The proportion of IDU-associated cases increased from 6.1% in 2014 to 14.5% in 2018. It is worth noting that the Tennessee EIP surveillance catchment area increased from 9 counties to 17 counties in 2017. In 2014, the proportion of candidemia cases with a history of IDU in the 9 original counties was 6.1% and increased to 15.3% in 2017. When the proportion was recalculated to include the additional 8 counties in 2017, the proportion of cases with a history of IDU was 14.0% (Table 1).

The remainder of the analysis includes 180 patients with candidemia (77 IDU-associated cases, 103 non-IDU cases) for which the secondary case report form was completed. Patients with IDU-associated candidemia were younger (median, 35 years; interquartile range [IQR], 28–43) than patients with non-IDU candidemia (61 years; IQR, 48–72; P < .001). Patients in both groups were predominantly white (IDU, 97%; non-IDU, 92%; P = .347). Although not statistically different, a greater proportion of IDU-associated cases occurred in women (55%) than in non-IDU cases (42%) (P = .162) (Table 2).

Hepatitis C infection was more common among patients with IDU-associated candidemia than those with non-IDU cases (77% vs 17%; P < .001). In contrast, patients with IDU-associated candidemia less frequently had underlying conditions traditionally associated with candidemia than patients with non-IDU cases, including diabetes (13% vs 35%; P < .001), abdominal surgery in the 3 months before diagnosis (3% vs 19%; P < .001), and receipt of TPN in the 14 days before diagnosis (5% vs 13%; P = .092). A similar proportion of patients with IDU-associated and non-IDU candidemia had CVCs placed during their hospitalization (52% vs 53%; P = .847), and, in particular, peripherally inserted central catheters (PICC lines) were commonly present during hospitalization (44% vs 34%; P = .165).

There was no statistically significant difference in onset classification between IDU-associated and non-IDU candidemia cases, but IDU-associated cases were more frequently classified as HACO than non-IDU cases (44% vs 32%; P = .096). We found that patients with IDU-associated candidemia had a median of 2 (range, 0–14) hospitalizations in the 6 months before diagnosis, compared with a median of 1 (range, 0–13; P = .022) for non-IDU cases. Endocarditis (39% vs 3%; P < .001) and septic emboli (36% vs 7%; P < .001) as end-organ manifestations of candidemia were more common among patients with IDU-associated infections than those without. During the 6 months before diagnosis, patients with IDU-associated candidemia were more frequently hospitalized for other infections than patients with non-IDU cases, including those involving cellulitis or skin abscesses (21% vs 6%; P = .002), endocarditis (27% vs 0%; P < .001), and septic emboli (21% vs 1%; P < .001).

Patients with non-IDU candidemia were more likely to have Candida glabrata infections (30% vs 17%; P = .041) and less likely to have Candida guilliermondii infections (0% vs 4%; P = .043) than patients with IDU-associated cases. The proportion of other Candida species was not significantly different among patients with IDU-associated and non-IDU candidemia. In the 6 months before candidemia, a higher proportion of patients with IDU-associated cases had other, non-Candida BSIs compared with patients with non-IDU cases (34% vs 16%; P = .004). They also more frequently had polymicrobial blood cultures at the time of their candidemia diagnosis or in the 6 months prior (31% vs 18%; P = .048).

In-hospital all-cause mortality was 4% among patients with IDU-associated candidemia and 17% among patients with non-IDU candidemia (P = .009). All-cause 30-day mortality was 8% in patients with IDU-associated candidemia and 25% in patients with non-IDU candidemia (P = .005). Ninety-day all-cause mortality was 14% in patients with IDU-associated candidemia and 36% in patients with non-IDU candidemia (P = .004).

Among patients with IDU-associated candidemia, opioids (81%) were the most commonly documented drug used by any route, followed by methamphetamines (22%), cannabinoids (10%), and benzodiazepines (10%), whether via injection or otherwise (Table 3). Of the patients with documented opioid use, 81% used prescription opioids, with 31% reportedly using oxymorphone specifically (Table 4). Medical records indicated that prescription opioids were commonly not used as prescribed and often indicated that patients were injecting their prescription opioid medications.

Based on 2016–2017 NSDUH estimates, the most recent years of available data, 82 000 Tennesseans aged >12 years (1.47%) ever injected drugs [23]. In 2017, approximately one quarter of all hepatitis C cases (23.8%) and nonfatal overdoses (24.3%) and one third of fatal overdoses (32.6%) occurred in the catchment area. We used the average of those 3 categories (26.9%) to estimate that 22 050 persons living in the catchment area used injection drugs, with 1477–6968 persons having done so in the previous year [22]. There were 28 IDU-associated candidemia cases in 2017, resulting in an estimated incidence of candidemia among PWID of 402–1895 cases/100 000 persons per year, which is 30–140 times the overall candidemia incidence in the catchment area. We determined that the overall annual incidence of candidemia was 13.5 cases/100 000 persons. The incidence of candidemia among the non-PWID population was 12.3 cases/100 000 persons.

Injection drug use-associated candidemia is increasingly common in East Tennessee; approximately 1 in 7 candidemia cases was associated with IDU in 2018, more than doubling the proportion since 2014. In 2017, the proportion of candidemia cases with a history of IDU in East Tennessee was higher than the national rate, at 14.0% compared with 9.7% nationwide according to EIP surveillance data [8]. The proportion of candidemia cases with a history of IDU among the 9 original counties in the surveillance catchment area increased ~9% from 2014 to 2017. Although the Tennessee EIP surveillance catchment area increased from 9 counties to 17 counties in 2017, this did not result in an increase in the proportion of cases with an IDU history. Therefore, this increase is unlikely to be attributed to this surveillance expansion, and the increasing rates are more likely attributable to the magnitude of the opioid crisis in Tennessee.

The estimated incidence of candidemia among PWID is at least 30 times the overall incidence in the study area. Even when compared with populations typically considered to have the highest candidemia incidence, specifically neonates (11.8/100 000) and people >65 years of age (55.7/100 000), incidence among PWID was at least 7 times higher [24, 25].

Among patients with IDU-associated candidemia, typical risk factors for candidemia, such as diabetes, malignancy, and receipt of TPN, were less commonly observed than in patients with non-IDU candidemia. Although the difference in frequency of underlying conditions between the 2 groups of patients is partially explained by the difference in age distribution, it also suggests that IDU contributes to an increased risk for candidemia in young people, because they lack candidemia-specific comorbidities. We were not surprised to find that a substantially higher proportion of patients with IDU-associated candidemia had hepatitis C than patients with non-IDU candidemia, as has been seen in other studies [26]. Given the known risk of viral hepatitis infection from IDU, the higher prevalence of hepatitis C in IDU-associated cases provides some reassurance that medical record abstraction appropriately categorized patients into the IDU-associated and non-IDU groups.

Previous descriptions of IDU-associated candidemia have focused on the risk of candidemia from the practice of injection. Nonsterile injection practices from inadequate skin disinfection, licking of needles before injecting (which can contaminate needles with Candida from the oral cavity), and use of juice from lemons harboring yeast have all been proposed as routes for introducing Candida into the bloodstream [26-28]. Based on these reports, we would expect that most IDU-associated candidemia would be community-associated and lack healthcare-associated risk factors. However, most IDU-associated candidemia cases in this analysis involved recent hospitalization, with an even greater number of hospitalizations in the preceding 90 days among PWID than non-PWID, many of which were for bacterial infections treated with antibiotics. Based on chart reviews and the data presented here, a common clinical course for a patient with IDU-associated candidemia included multiple previous admissions for invasive bacterial (sometimes polymicrobial) infections, continuation of intravenous antibiotic therapy through a PICC after discharge, and readmission for candidemia with a PICC still present. These data suggest that classic healthcare-associated risk factors for candidemia (ie, CVCs and previous antibiotics) may contribute more to IDU-associated candidemia than previously expected. Outpatient PICCs may play a role in IDU-associated candidemia by allowing Candida contamination through poor PICC maintenance or illicit use of drugs through this device. Further study of this association is warranted. Outpatient parenteral antimicrobial therapy using PICC lines has been proposed as an option for treatment of infections in PWID, but Infectious Diseases Society of America Guidelines provide no recommendations on this practice [29, 30]. Given the potential risk of candidemia and other infections, physicians should consider ways to reduce the risk of PICC-associated infections in PWID, possibly including periodic assessments of need for PICCs and enhanced safe insertion and maintenance practices.

A significantly higher proportion of IDU-associated cases involved end-organ disease when compared with non-IDU cases. A study of patients with invasive methicillin-resistant S aureus similarly found higher rates of septic embolism, endocarditis, and osteomyelitis among patients with a history of IDU than patients without an IDU history [31]. The morbidity and mortality related to end-organ disease is significant; for example, studies have found that fungal endocarditis results in >50% mortality in some populations and requires advanced medical therapy that includes combination valvular surgery and antifungal treatment [32, 33]. Fungal endocarditis is often associated with a history of IDU and the presence of other valve disease, including previous bacterial endocarditis [33, 34], which was frequently documented in the medical records we reviewed.

Although our study found that the mortality was lower for IDU-associated candidemia than non-IDU cases, the rate was still high for a relatively young population. Similarly high infection-related mortality is seen due to IDU-associated methicillin-susceptible S aureus, with 11% mortality even with the receipt of empirical therapy [35].

Our study has several notable limitations. The number of IDU-associated candidemia cases is likely an underestimate. Patients may not have disclosed all IDU history, IDU may not be accurately represented in inpatient medical records, and PWID may not have had the resources available to seek medical care. All of these factors would lead to underestimation of candidemia incidence among PWID. It is possible that the increase in the proportion of IDU-associated cases was because of the growing attention and awareness of the opioid crises in more recent years and providers more diligently recording IDU history in medical records. The estimated incidence calculations used a state-level estimate of self-reported data to determine a count of Tennesseans who inject drugs and made a number of assumptions to calculate county-level PWID denominators. However, there are no comprehensive data sources for number of PWID in a county. We used a range of estimates for lifetime PWID that may have injected drugs in the past year and an average of underlying conditions associated with PWID to derive our estimates in the absence of more robust data. The use of the input of 6.7%–31.6% of lifetime PWID injecting drugs in the past year is likely a conservative range for estimating incidence, because the only available estimates come from self-reported data. Finally, our analysis examined only those cases of candidemia that were detected through the EIP surveillance system in East Tennessee and may not be representative of the incidence, clinical outcomes, and mortality seen in other regions, particularly because drug use patterns can be highly regional.

These results add to the body of work on infectious disease complications associated with IDU and the opioid crisis. Candidemia in PWID is not an isolated incident. It frequently occurs along a clinical course that includes several previous and subsequent infections and hospitalizations, followed by death within 90 days among 1 in 7 patients, often related to other infections or overdose. Each encounter with a healthcare facility could be seen as a potential opportunity to link patients to much needed substance use treatment and harm reduction services. To improve these hospital-based efforts, training programs for medical personnel could emphasize methods for identifying and treating patients with opioid use disorder (OUD), and hospitals could implement standard inpatient protocols for the initiation of medication-assisted treatment for OUD and subsequent linkage to community-based treatment [36].

We have documented IDU as an increasingly common exposure among patients with candidemia and that incidence is much higher among PWID than the general population. Clinicians who diagnose candidemia in patients without typical risk factors should consider IDU as a possible risk factor. These patients should be screened for viral hepatitis, HIV, and other infectious diseases associated with IDU. Hospitalization for candidemia treatment could be leveraged as an opportunity to link patients to appropriate substance use treatment, mental health, and harm reduction services [37-39]. In addition, candidemia should be considered a potential adverse event when weighing the risks and benefits of discharging patients with a PICC line.