Infectious Diseases Society of America (IDSA) guidelines recommend routine ophthalmic screening for patients with systemic candidiasis.1 Primary teams will frequently request ophthalmology consultations to rule out endophthalmitis in inpatients with systemic fungal or bacterial infection.2,3 These requests are reported behind only “decreased vision” and “red eyes” as the most frequent reason for inpatient consultation.2
Despite this, there are more inpatients with infection than patients who actually receive consultations. Scant literature exists regarding characteristics that prompt a request or deferral of consultation. One study on fungemic inpatients found that mortality and hospice transfer affected consultation rates, but further analyses were not performed.4
Similarly, there are more patients who receive consultations than patients who have endophthalmitis. Many retrospective chart reviews have described clinical characteristics of patients who develop endogenous endophthalmitis.5–7 However, fewer studies have rigorously calculated the relative likelihood of developing endophthalmitis based on patient or hospitalization characteristics. Among the published analyses, comorbidities such as length of stay, intensive care unit (ICU) stay, total parenteral nutrition (TPN), and altered mental status were found to be associated.8–10
Identifying risk factors for endogenous endophthalmitis and comparing these characteristics to those of patients who are selected to receive screening ophthalmology consultations will better guide clinical decision-making. A multicenter, prospective study of sufficient size would be daunting in scope; however, structured electronic medical record (EMR) data offers an opportunity to better study rare conditions such as endophthalmitis.
In this study, we use EMR data spanning more than 5 years at a tertiary care hospital and encompassing more than 80,000 patients with more than 134,000 clinical encounters to examine a broad spectrum of risk factors for endophthalmitis among inpatients with systemic infection and identify characteristics that predispose these patients to receiving an ophthalmology consultation.
Patients and Methods
This was a retrospective cohort study at Stanford Medical Center. Inpatient encounters between January 1, 2010, and December 31, 2014, were identified using Stanford Translational Research Integrated Database Environment (STRIDE). STRIDE is a research and development project at Stanford University to create a standards-based informatics platform supporting clinical and translational research.11 Institutional review board approval was obtained, and practices adhered to the Declaration of Helsinki. Data were managed in a HIPAA-compliant fashion.
Patients were included if they had systemic bacterial or fungal infection (Figure 1). The bacterial cohort (n = 4,592) only contained patients with positive bacterial blood cultures who also received intravenous (IV) antibacterials due to the frequency of skin contaminants resulting in false-positive cultures.12 The fungal culture cohort (n = 1,905) had positive fungal cultures from any source; we also evaluated a positive fungal blood culture subset (fungal blood culture cohort, n = 332). The antifungal cohort (n = 5,023) contained those who received IV antifungals. We distinguished the fungal culture cohort and antifungal cohort, as ophthalmologists can be consulted for patients who are not fungal culture-positive but who receive antifungals due to high clinical suspicion.5 Finally, we combined the antifungal cohort and fungal culture cohort (total fungal cohort, n = 5,963) for certain analyses. Patients could be classified into multiple cohorts, with 18.2% (1,734 / 9,527) part of multiple cohorts.
Sample selection strategy.
We identified 9,527 inpatient encounters out of more than 134,000 total during the 5-year study period. International Classification of Disease, 9th Edition, Clinical Modification (ICD-9-CM) codes with supplemental manual review identified patients with endophthalmitis. Miscoded cases (such as for cotton-wool spots and leukemic infiltrates) were excluded. Inclusion criteria for true endophthalmitis included chorioretinal lesions and/or vitritis in the setting of compatible medical history.
ICD-9-CM codes (Supplemental Table A available at www.healio.com/OSLIRetina) identified comorbidities suspected as risk factors for endophthalmitis (eg, “bacterial meningitis”). Selection of these variables was guided by prior studies.8,9 Other variables, including demographics, admission date, admitting service, length of stay, presence of ophthalmology or infectious disease (ID) consult, ICU admission, English as a second language, TPN, Foley catheter, or steroid treatment were obtained from STRIDE. Intravenous drug usage, although of interest, was not captured as it was unable to be accurately defined using STRIDE or ICD-9-CM coding.
ICD-9-CM Codes Used
Stepwise regression analysis was performed using STATA version 14.1 (StataCorp, College Station, TX). Preliminary univariate analyses identified variables associated with (1) receiving ophthalmology consultation or (2) developing endophthalmitis (Supplemental Table B available at www.healio.com/OSLIRetina). A P value of less than .01 was set as significant due to the large number of comparisons.
Univariate Predictors of Ophthalmology Consult and Endogenous Endophthalmitis
Multivariable logistic regression was performed using methods described previously.8 For ophthalmology consultation, we evaluated demographic variables and the 18 most clinically or statistically significant univariate predictors. For endophthalmitis, few univariate variables were significant. For comparison we used similar variables in our multivariable regression of endophthalmitis as we did for consultation. No patient with bacterial meningitis or HIV had endophthalmitis; these were excluded from the final endophthalmitis analysis. All patients with endophthalmitis received infectious disease consultation; this variable was excluded from final endophthalmitis analysis due to collinearity.
Interaction effects existed between fungal culture (any source or blood source) and fungal medication; we, therefore, included interaction effect analyses in final models. Because fungal blood culture was more significant than fungal culture from any source in univariate, we only used fungal blood culture in our final analysis.
Incidence and Baseline Characteristics
Twenty-five cases of endophthalmitis were identified — a 0.3% incidence over 5 years among all patients (25/9,527) and 3.1% among patients with ophthalmology consultations (25/803). The 5-year incidence for fungal endophthalmitis was 0.3% (15 / 5,963 from the total fungal cohort) and for bacterial endophthalmitis was 0.2% (8/4,592 from the bacterial cohort).
Ophthalmology consultations were more frequent than endophthalmitis. During a 5-year span, 8.4% (803/9,527) received consultations. The 5-year incidence for consultation was 11.8% among fungal patients (704/5,963 from the total fungal cohort) and 6.6% among bacterial patients (301/4,592 from the bacterial cohort). All patients with endophthalmitis received ophthalmology consultations.
Baseline characteristics are reported in Table 1 for all patients and for the subset with endophthalmitis. Patients in both groups were predominantly Caucasian, more likely to be admitted under internal medicine, and covered a broad age range.
Demographics and Clinical Characteristics of Inpatients With Bacterial or Fungal Infection
Clinical details for the 25 endophthalmitis cases are presented in Table 2. The causative organism was fungal in 15 patients, bacterial in eight, and unclear in two (cases 24 and 25). Candida species and methicillin-resistant Staphylococcus aureus were the most common fungal and bacterial causes.
Clinical Characteristics of Positive Endophthalmitis Cases
Four out of 25 patients had negative blood cultures, instead having positive cultures from other sources or no positivity at all (Cases 1, 2, 3, and 25). Of the 20 patients with adequate mentation, six were asymptomatic (Cases 4, 5, 6, 9, 12, and 13). All asymptomatic patients had fungal endophthalmitis. Patients with vitritis were symptomatic and patients without were asymptomatic, except in cases when patients had confounding ocular conditions (eg, leukemic retinopathy; Cases 11, 15, and 23).
Risk Factors for Endophthalmitis
In univariate analyses (Supplemental Table B), few comorbidities significantly affected odds of endophthalmitis. Most significant were endocarditis, fungal blood culture, and ages 45 to 54 years (compared with 65 years and older). These variables remained significant in multivariable analysis (Table 3), with positive fungal blood cultures having an over 13-fold increase in odds.
Clinical Predictors for Ophthalmology Consultation or Endophthalmitis
Risk Factors for Ophthalmology Consultation
In univariate analyses (Supplemental Table B), multiple factors were associated with ophthalmology consultation, most significant being positive fungal blood culture (over 22-fold odds increase), followed by longer length of stay (31 or more days compared with 0 to 4 days), fungal meningitis, HIV, and infectious disease consultation.
In multivariable analysis (Table 3), positive fungal blood cultures had a more than 20-fold odds increase, followed by HIV, infectious disease consultation, and longer length of stay (31 or more days compared with 0 to 4 days). Other significant variables included younger age, intravenous (IV) antifungal usage, endocarditis, and diabetes with eye manifestations. Bacterial and fungal meningitis were also positively associated with consultations but did not meet our P value of less than .01 threshold for statistical significance.
In contrast, TPN usage was associated with a 26% decrease in odds (P = .009). ICU stay (P = .015) and surgery admitting service (P = .034) trended toward decreased odds, although they did not meet our P value of less than .01 statistical significance threshold. It should be noted that endocarditis, although significant for both ophthalmology consultation and endophthalmitis, only had 48% higher odds for ophthalmology consultation (95% CI, 1.20–1.84), compared with a more than 7-fold odds increase for endophthalmitis (95% CI, 3.11–17.5).
Endogenous endophthalmitis frequently results in poor outcomes.6,13 Identifying patients most at risk may prevent devastating visual decline. However, positive findings are rare, and recent literature suggests that routine consultation may not constitute efficient resource utilization given current low endophthalmitis rates and earlier initiation of antimicrobial treatment.4,14–16 Understanding predictors that affect endophthalmitis development and optimizing consultation practices will be important for improving utilization.
In this EMR analysis of a tertiary care university hospital, we identified 9,527 inpatients over 5 years who developed bacteremia, a fungal infection, or who received IV antifungals. Twenty-five (0.3%) of these patients developed endophthalmitis: 15 (0.3%) from the total fungal cohort and eight (0.2%) from the bacterial cohort. Our study is unique in that our baseline cohort included all patients with signs of systemic infection, in contrast to prior studies that only included patients who had ophthalmology consultation requests.5 There was a 3.1% incidence (25/803) in the subset of patients for who received ophthalmology consultations.
Candida and S. aureus (predominantly methicillin-resistant) were the most common cause of fungal and bacterial endophthalmitis, similar to findings in prior studies.17,18Aspergillus and Cryptococcus comprised two cases each. Finally, we identified isolated cases of Trichosporon, Serratia, and Streptococcus, which have been described before rarely.19–21
During the same 5-year period, 8.4% of patients received an ophthalmology consultation: 11.8% and 6.6% of patients in the total fungal cohort and bacterial cohort, respectively. The consultation rate for fungal patients differs from prior reports, which have suggested that up to 68.7% of fungemic patients receive ophthalmology consultations.4 This may be because we used a broad definition for fungal patients, including those with positive non-blood cultures or who received antifungals without positive cultures.
Ophthalmology consultation is often prompted by positive blood cultures or presence of symptoms. However, four of our 25 endophthalmitis patients had negative blood cultures, instead having growth from other sources, or no growth at all. This is consistent with prior reports22 and lends credibility to the practice of consulting ophthalmologists if there is a strong suspicion of endophthalmitis, even when cultures are negative.5,22 Furthermore, of the 20 patients with adequate mentation, six had positive findings despite being asymptomatic. Thus, despite prior proposals of symptoms as a way to screen for endophthalmitis,14 our results suggest that it may be necessary to screen patients even when asymptomatic, echoing recommendations of other studies.5,8 It is worth noting that all six asymptomatic patients had fungal endophthalmitis. All patients with adequate mental status and bacterial endophthalmitis were symptomatic. Although physicians could limit consultations for bacteremic patients with adequate mentation to only evaluate those with symptoms, this approach may not be prudent.
Casting a wide net in consultation requests will inevitably yield many negative exams, however. In order to identify the 25 endophthalmitis patients in our analysis, 803 patients received consultations — 0.3% versus 8.4% of our total cohort. The number needed to screen to detect one case of endophthalmitis would be 32 (25/803). This prompts questions regarding finding better ways to identify at-risk patients.
Research is sparse regarding rationale for ophthalmology consultations. Ghodsara et al. found no difference in age or gender of fungemic patients regarding consultation rate. However, consultation was less likely when a patient was transferred to hospice or passed away during admission.4
In our study, we found the most significant factors associated with ophthalmology consultation to be fungal blood culture, HIV, infectious disease consultation, and longer length of stay. This correlates with clinical experience; patients with positive blood fungal cultures or who receive an infectious disease consult are more likely to receive an ophthalmology consult because of IDSA guidelines. Patients with HIV may be more likely to have ophthalmology consultations due to immunocompromised status. Those with longer hospitalizations will have a higher risk of developing systemic infection, as well as prolonged opportunity for consultation. Interestingly, receiving TPN had lower odds of consultation, and being admitted under a surgical service and having an ICU stay trended toward decreased odds. Perhaps in these perioperative or critically ill patients, attention is focused on systemic illness, leading to deferral of ophthalmic screening exams.
Although the rarity of endophthalmitis limited our ability to identify significant endophthalmitis risk factors, we found substantial associations with fungal blood culture and endocarditis. These results corroborate previous studies.9 It is reassuring that positive fungal blood cultures are associated with both endophthalmitis and ophthalmology consultation. However, endocarditis was not as strongly associated with receiving a consultation (48% increase in odds) as with developing endophthalmitis (more than a seven-fold increase in odds). Perhaps inpatients with endocarditis should receive additional attention for screening ophthalmology consultations.
Importantly, our results only reflect findings from a single hospital system and may not be generalizable to all settings. Due to small sample size, the incidence of endophthalmitis was low and thus certain variables (such as bacterial meningitis and HIV) were excluded from final analysis. Another major limitation is use of administrative ICD-9-CM codes. Studies conflict regarding the reliability of ICD-9-CM coding, with some suggesting that these codes are accurate, and others arguing that they are less useful for esoteric diagnoses.23,24 Nonetheless, despite the inherent weaknesses, the ability to use clinical informatics to extract data does allow a larger cohort and better generalizability. In this analysis, we additionally relied on non-ICD-9-CM elements directly extracted from the medical record and performed a manual review of endophthalmitis cases.
In conclusion, presence of positive fungal blood cultures was the most significant predictor for developing endogenous endophthalmitis and for receiving an ophthalmology consultation. The presence of endocarditis, although strongly associated with endophthalmitis, was less strongly associated with consultations. Increased attention to consultations for patients with endocarditis could have value in identification of at-risk patients. Moreover, the presence or absence of ophthalmic symptoms or of positive blood cultures may not be an accurate way to judge need for consultation, given that some endophthalmitis patients lacked one or both of these. Further research is needed to comprehensively evaluate potential clinical risk factors to enable more targeted consultation for this rare condition — potentially using tools such as large clinical data registries to achieve sufficient sample size.