Prevalence of Drug Fever among Cases of Nosocomial Fever: A Systematic Review and Meta-analysis

Authors: Hidehiro Someko, Yuuji Okazaki, Yasutaka Kuniyoshi, Akira Yoshida, Keisuke Baba, Atsuhiro Ijiri, Yasushi Tsujimoto

Abstract

Objective

Drug fever is defined as a fever that temporally coincides with the start of a culprit drug and disappears after discontinuation of the drug. It is a common cause of nosocomial fever, which refers to a fever that develops beyond the first 48 h after hospital admission. However, the exact prevalence of drug fever among cases of nosocomial fever is unclear, as is the variation in prevalence depending on the clinical setting and most common causative drugs.

Methods

PubMed MEDLINE, Dialog EMBASE, Cochrane Central Register of Controlled Trials, World Health Organization International Clinical Trials Registry Platform, and ClinicalTrials.gov were systematically searched. Studies that reported the prevalence of drug fever in patients with nosocomial fever were included. Two of the four reviewers conducted independent assessments of the inclusion, data extraction, and quality. Pooled adjusted odds ratios were generated using a random-effects model and presented with 95% confidence intervals (CIs).

Results

Fifteen meta-analysis from 15 studies were included. Ten studies did not report the definition of drug fever or excluded febrile patients who were admitted to the hospital within 24-48 h. The pooled prevalence of drug fever among cases of nosocomial fever was 3.0% (95% CI, 0.6-6.8%), which was largely consistent across the settings, except for at oriental medicine hospital. Only four studies reported the causative agents, and antibiotics were the most frequently reported.

Conclusions

The prevalence of drug fever is low in patients with nosocomial fever. Clinicians should recognize that drug fever is a diagnosis of exclusion, even in cases of nosocomial fever.

Introduction

Nosocomial fever is defined as a fever occurring 24-48 h after hospitalization (1). The incidence of nosocomial fever in hospitalized patients is 2-29% (1-3). Although infectious causes were reported to be the most frequent, accounting for 37-74% of cases, non-infectious causes are not negligible, accounting for 3-52% of cases (1). An accurate distinction between infectious and non-infectious causes is important, as the unnecessary administration of antibiotics is costly and may lead to frequent adverse drug reactions (ADRs) (4,5).

Drug fever is an important differential diagnosis for patients with nosocomial fever. A misdiagnosis of drug fever may lead to inappropriate or potentially harmful diagnostic or therapeutic interventions, resulting in prolonged hospitalization and additional medical costs. Although the prevalence of drug fever among cases of nosocomial fever has been reported to vary depending on patient characteristics and settings, no comprehensive studies have been conducted on this topic (1). An overview of the prevalence of drug fever may reduce misdiagnoses by clarifying the probability of drug fever in hospitalized patients.

We conducted a systematic review and meta-analysis of the prevalence of drug fever among cases of nosocomial fever to clarify (i) the pooled estimates of the prevalence of drug fever among cases of nosocomial fever, (ii) the difference in the prevalence of drug fever among cases of nosocomial fever between settings, and (iii) the causative agents of drug fever among cases of nosocomial fever.

Materials and Methods

We partially followed the Cochrane Handbook (6) and Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) 2020 (7) to conduct this systematic review and meta-analysis (Supplementary material 1). The review protocol was registered at osf (https://osf.io/n9trx/).

Inclusion criteria of the articles for the review

We included all observational studies that reported the incidence of drug fever in patients with nosocomial fever. We did not restrict the publication date, status (full publication, conference abstract, or unpublished data), or language. Case reports, case series, case-control studies, and reviews were excluded.

We included all patients with nosocomial fever, regardless of comorbidity, age, sex, or race. We accepted any setting, including the intensive care unit (ICU), internal medicine ward, postsurgery, and pediatrics.

Nosocomial fever was defined as a fever occurring ≥48 h after hospital admission in afebrile patients on admission or that developed ≥7 days after the initial fever resolution in patients who were admitted with a fever (1). A fever was defined as a body temperature ≥37.8°C (8,9), measured by either invasive or non-invasive methods. We also adopted the definition of nosocomial fever adopted by the authors of the included studies.

Outcomes

The primary outcomes were (i) prevalence of drug fever in patients with nosocomial fever, (ii) prevalence of drug fever in different clinical settings, and (iii) types of causative drugs for drug fever. Drug fever was defined as a febrile episode that temporally coincided with drug administration and disappeared (defervesced) after withdrawal of the offending drug, without other causes identified through careful workup, including history taking, a physical examination, and laboratory investigations (10).

Secondary outcomes included the proportion of patients with skin rash, eosinophilia, or both among patients with drug fever and nosocomial fever.

Search methods

PubMed MEDLINE, Dialog EMBASE, Cochrane Central Register of Controlled Trials, World Health Organization International Clinical Trials Registry Platform (ICTRP) Search Portal, and ClinicalTrials.gov were searched from inception to November 23, 2022, using keywords related to nosocomial fever and drug fever. The detailed search strategy is shown in Supplementary material 2.

Data collection and analyses

Two pairs of four independent authors (HS, AY, KB, and AI) screened the titles and abstracts of each study, which were identified using the search strategy and evaluated for eligibility. The full texts of the selected studies were collected and independently evaluated for eligibility by the authors. The original authors were contacted for studies with only abstracts available that had unclear eligibility. Any disagreements between reviewers were resolved through discussion. Disagreements were discussed with a third independent reviewer if necessary.

Two pairs of authors (HS, AY, KB, and AI) independently performed data extraction from the eligible studies. The following data were study country, observational time, study design, clinical setting, inclusion and exclusion criteria, definition of nosocomial fever, definition of drug fever, demographics of participants, sample size, number of patients diagnosed with drug fever, and the type and name of the causative drugs. Drugs were classified according to the pharmacological-therapeutic classification system of the American Hospital Formulary Service (11). Any discrepancies between authors were resolved through discussion. We discussed these issues with a third independent reviewer if necessary. The original authors were contacted and asked for study details, as needed.

Assessment of the risk of bias in included studies

Two pairs of 4 authors (HS, AY, KB, and AI) independently assessed the risk of bias for each study using the Joanna Briggs Institute Prevalence Critical Appraisal Tool (12,13). Any disagreements between reviewers were resolved through discussion. Disagreements were discussed with a third independent reviewer if necessary. We defined the quality score for each study as the proportion of “Yes” answers to the total number of questions evaluated. Quality scores were categorized as low if the proportion was ＜50%, moderate if it was 50-80%, and high if it was ＞80% (14).

Meta-analyses

We synthesized the data of the prevalence of drug fever in patients with nosocomial fever using the R 4.2.3 software program with the “meta” (15) and “metafor” packages (16) and combined proportion with 95% confidence intervals (CIs) and 95% prediction intervals (PIs). To determine the CIs for the results of the individual studies, we used the inverse variance-weighted random-effects model with the DerSimonian-Laird estimator. This approach allowed us to estimate between-study variance and construct a normal approximation interval based on summary measures. To ensure the stability of the variances, we used the Freeman-Tukey double arcsine transformation (17). We performed a meta-analysis of all included studies regardless of the clinical setting, as whether or not the prevalence of drug fever in patients with nosocomial fever differs among clinical settings is unclear. We added a column for each type of causative drug of drug fever in patients with drug fever to the table of characteristics of the included studies.

Assessing heterogeneity

Heterogeneity was visually assessed using forest plots. I^2^ statistics were computed using the Cochrane chi-squared test (Q-test). We also calculated the τ^2^ statistics for the PI.

Subgroup analyses

To evaluate the clinical heterogeneity of the included studies, we performed a meta-analysis by dividing the included studies into several different clinical settings (all departments, internal medicine, ICUs). We defined the setting of a study as “all departments” if the study had no restrictions for departments, including patients.

Sensitivity analyses

To confirm the robustness of the pooled prevalence of drug fever in patients with nosocomial fever, we performed the following sensitivity i) excluding studies that were evaluated to have low-quality scores, ii) excluding studies that were considered highly heterogeneous because the exposure was substantially different from all other studies, and iii) excluding studies that did not clearly define drug fever.

Differences between the protocol and review

We performed a subgroup analysis by dividing the included studies by patient age (＜65 or ≥65 years old). However, grouping the studies by patient age resulted in similar groupings in clinical settings (e.g. pediatrics). Therefore, we ultimately did not include a subgroup analysis according to age. We planned to perform a sensitivity analysis by excluding studies without a “wash out” period of community-acquired febrile illnesses. However, little heterogeneity was observed between the studies with and without a “wash out” period. Therefore, we ultimately did not perform a sensitivity analysis. We added a sensitivity analysis by excluding studies that did not clearly define drug fever because these studies may have overestimated the prevalence of drug fever by including other diseases similar to drug fever (e.g. gout, pseudogout, deep venous thrombosis, aspiration pneumonitis).

Results

We retrieved 1,428 articles using our search strategy (Fig. 1). Following the removal of duplicates and initial screening process, 36 full-text articles were evaluated for eligibility. Of these, 22 studies (23 reports) did not meet the inclusion criteria and were excluded (Supplementary material 3), resulting in 12 studies (13 reports) satisfying the inclusion criteria. Three additional studies were identified through a citation search. Consequently, 15 studies (16 reports) were included in our meta-analysis (2,18-31). Note that studies here refer to the individual units of research, while reports encompass any written documents or publications that provide information about studies. Reports include journal articles, conference papers, theses, dissertations, and any other format in which the research findings are reported.

The analysis included 2,735 nosocomial fever episodes (2,733 patients). The main characteristics of the included studies are presented in Table 1 and Supplementary material 4. Of the 15 included studies, 4 did not restrict the department (21,22,25,28), 3 were conducted in internal medicine wards (2,23,26), and 5 were conducted in the ICU (24,27,29-31). Other settings included pediatric (19), obstetric and gynecological (18), and oriental medicine hospitals (20). Three studies focused on patients with specific conditions, such as post-meningitis treatment (19), spinal cord injury (24), and post-liver transplantation (29).

One study each was categorized as of a high quality (20) or low quality (18). All others were of a moderate quality (Table 2). A low to moderate quality was observed mainly due to the diversity of the clinical settings, definition of nosocomial fever, and insufficient reporting of the definition of drug fever. The definition of nosocomial fever used in the present study was adopted from five previous studies (2,20,23,25,27). The definition of drug fever was only reported in five studies (2,19,20,25,27).

The pooled prevalence of drug fever among cases of nosocomial fever was 3.0% (95% CI, 0.6-6.8%; 95% PI, 0-26.7%) (Fig. 2). When restricted to studies with patients admitted to all departments, the pooled prevalence was 5.8% (95% CI, 1.9-11.3%; 95% PI, 0-40.3%) (21,22,25,28). When the analysis focused on studies conducted in internal medicine services, the pooled prevalence was 2.0% (95% CI, 0-6.9%; 95% PI, 0-100%) (2,23,26). In studies involving patients admitted to the ICU, the pooled prevalence was 0.4% (95% CI, 0-3.0%; 95% PI, 0-16.3%) (24,27,29-31). None of the studies reported the number of patients with skin rash, eosinophilia, or both among those with drug fever.

Four studies reported the name or type of the causative drug for drug fever (2,18,20,22). Two studies reported antibiotic use as the main cause (18,22) In one study, the causative drugs were herbal medicines and antibiotics (20), whereas in another study, cimetidine was identified as the causative agent (2).

The sensitivity analysis, excluding a low-quality study (18), demonstrated that the pooled prevalence of drug fever among cases of nosocomial fever was 3.1% (95% CI, 0.4-7.5%; 95% PI, 0-30.3%). A second sensitivity analysis, with the exclusion of a study on oriental medicine hospitals and thus considered to be in very different settings (20), demonstrated that the pooled prevalence of drug fever among cases of nosocomial fever was 2.3% (95% CI, 0.6-4.6%; 95% PI, 0-15.0%). A third sensitivity analysis that incorporated only studies that defined drug fever clearly (2,19,20,25,27) demonstrated that the pooled prevalence of drug fever among cases of nosocomial fever was 5.4% (95% CI, 0.6-13.8%; 95% PI, 0%-49.9%) (Supplementary material 5). Thus, the pooled prevalence of drug fever in patients with nosocomial fever did not change substantially throughout the sensitivity analysis.

Discussion

This systematic review and meta-analysis revealed that the pooled prevalence of drug fever in patients with nosocomial fever is approximately 3%. However, the methodological quality of the included studies was low to moderate because of the selection of the patient population, inconsistency in the definition of nosocomial fever in our review, and the lack of reporting on the definition of drug fever. The prevalence was largely consistent across settings, except for the oriental medicine hospital. Although the causative agents of drug fever have rarely been reported, several studies have suggested that antibiotics are the most common cause of drug fever among cases of nosocomial fever.

The pooled prevalence of drug fever in patients with nosocomial fever was lower than that previously reported. In a narrative review of drug fever, the prevalence of nosocomial fever in patients was approximately 10% (32). The diagnosis of drug fever is generally established by observing the resolution of a febrile episode after cessation of suspected medication. Diagnosing drug fever within a short period after a patient develops fever is difficult. Given the low prevalence of drug fever among cases of nosocomial fever, even in cases where drug fever is highly suspected based on signs and symptoms, the exclusion of other possible differential diagnoses may be important for clinicians.

The pooled prevalence of drug fever and nosocomial fever observed in the present review should be interpreted with caution for several reasons. First, drug fever was only explicitly defined in five studies (2,19,20,25,27). The definition of drug fever was similar in these studies; however, the definitions of drug fever in a previous study of drug fever varied depending on whether or not the following conditions were i) complications associated with drug administration in drug fever (33), ii) febrile responses with skin manifestations resulting from drug hypersensitivity from drug fever, and iii) drug-induced hyperthermia (e.g. neuroleptic malignant syndrome and malignant hyperthermia). Thus, the reported prevalence of drug fever in each study was consistent, suggesting that the adopted definitions of drug fever were similar (10). Our sensitivity analysis with the exclusion of studies that did not clearly define drug fever showed a slightly higher prevalence of drug fever among cases of nosocomial fever; however, this may be because one of the included studies in the sensitivity analysis had a markedly higher prevalence than others (20). Second, drug fever may have been under-diagnosed in the included studies. Drug fever is generally considered to be a diagnosis of exclusion. If a clinician considers another differential diagnosis, the patient is more likely to be diagnosed with this disease. For example, when a patient's fever reemerges despite antibiotic administration, clinicians generally perform additional workups, including laboratory testing, urinalysis, and chest radiography. If a urinalysis reveals pyuria and bacteriuria, clinicians may diagnose the patient with a new urinary tract infection (UTI) and change antibiotics to target the suspected UTI pathogen. If a patient's fever resolves after a change in antibiotics, clinicians should consider confirming the diagnosis of UTI. However, the possibility of drug fever cannot be excluded. Such a possibility of a misdiagnosis or misclassification could lead to an underestimation of the prevalence of drug fever among cases of nosocomial fever, even when clinicians' perception of the definition of drug fever is consistent with our review. In cases where the diagnosis is unclear, readministration of the drug (rechallenge) may help clarify the diagnosis; however, this approach was not performed in any of the included studies. Further studies with a clear definition of drug fever and efforts to correct its diagnosis are necessary to determine the exact prevalence of drug fever among patients with nosocomial fever.

Only slight differences were found in the prevalence of drug fever among cases of nosocomial fever between clinical settings, except for oriental medicine hospitals. There are two possible reasons for this. First, no risk factors for drug fever specific to patient characteristics have been identified, except for genetic polymorphisms (10,34). The distribution of genetic polymorphisms associated with drug fever may not differ significantly among clinical settings. Second, the prevalence of drug use, which often causes a fever, may be consistent across different settings. A systematic review and meta-analysis of the prevalence of ADRs in hospitalized patients reported that diuretics, antibiotics, and antithrombotics are common classes of drugs that cause ADRs in hospitalized patients (35). Of these, antimicrobials are associated with drug fever. Antibiotics are widely used in hospitalized patients regardless of the clinical setting (36). This may explain the similarity in the prevalence of drug fever among cases of nosocomial fever between different clinical settings, except for oriental medicine hospitals, where herbal medicines are frequently used.

Antibiotics may be associated with drug fever in nosocomial fever patients. Three of the four studies included antibiotics as causative agents. This may be because antibiotics are associated with a fever and are widely prescribed to hospitalized patients. Suspicion of drug fever may be advisable for clinicians in patients with nosocomial fever on antibiotics. Further studies are warranted to investigate whether or not the pre-test probability of drug fever is higher in patients with nosocomial fever who are taking antibiotics than in those who are not.

Several limitations associated with the present study warrant mention. First, our search strategy included keywords related to the outcomes (drug fever). One standard practice suggested for conducting systematic reviews is to avoid including keywords for outcomes in the search strategy (37). We may have missed studies in which the prevalence of drug fever among cases of nosocomial fever was zero and thus was not explicitly reported. To minimize this bias, our search strategy incorporated keywords with broader concepts than drug fever (non-infectious cause). In addition, we used a combination of citation searches. We included three studies in which the prevalence of drug fever in patients with nosocomial fever was zero. Second, we did not find any studies on general surgery or urology. Although several of the included studies were conducted in all hospital departments, the results may not be applicable to specific settings. Third, all included studies were conducted in acute-care settings. Finally, the heterogeneity was large in our study, although it is a common observation in meta-analyses of prevalence. This may have been due to variations in the definitions of nosocomial fever and drug fever along with disparities in the demographics of the participants. Nevertheless, these factors appear to exert a minimal overall impact, prompting us to conduct a comprehensive meta-analysis. Furthermore, the outcomes of our sensitivity analysis consistently supported our decision, reinforcing the validity of our approach. We believe that this meta-analysis of all studies is justifiable, but we must remain mindful of this inherent heterogeneity. The results may differ between rehabilitation and chronic-care hospitals. Further studies are therefore required to investigate whether or not our results can be replicated in different clinical settings.

In conclusion, available evidence suggests that the prevalence of drug fever among cases of nosocomial fever is low. Drug fever should be considered as a diagnosis of exclusion from nosocomial fever. However, the true prevalence remains uncertain, considering the heterogeneity of the clinical settings and the adopted definitions of nosocomial fever and drug fever in the included studies. Clinicians should be aware of the precise definition of drug fever when diagnosing suspected cases of drug fever. Further studies are warranted to determine the exact prevalence of drug fever among patients with nosocomial fever in different clinical settings.

Author's disclosure of potential Conflicts of Interest (COI).

Yasushi Tsujimoto: Grants, Pfizer Health Research Foundation.

Supplementary Material