According to internal review board–approved parent cohort protocols, participants were enrolled upon hospitalization. We collected demographic, symptom, examination finding, and laboratory data on standardized forms during hospitalization and at 1 month after enrollment (Appendix) (13,17; P.W. Blair et al., unpub. data, https://www.medrxiv.org/content/10.1101/2023.09.14.23295526v1). We collected acute blood samples at enrollment and convalescent samples at 1 month. We determined survival during in-person visits or telephone calls during a 1–3-month period after hospitalization in the acute febrile illness (AFI) cohort and a 12-month period in the sepsis cohort.

We used acute-phase serology from archived samples collected at parent study enrollment to determine seroprevalence in the sepsis (n = 311) and AFI (n = 122) cohorts. We conducted IgG IFA by using commercial slides (spotted fever group rickettsia, Rickettsia conorii Malish 7 strain; typhus group rickettsia, Rickettsia typhi Wilmington strain) (Biocell Diagnostics Inc.). We screened serum samples collected at time of hospitalization (acute phase) and at 1-month follow-up (convalescent phase) at a dilution of 1:64 and titrated to 1:65,536 (17). We considered a sample seropositive at a threshold titer of >128. We used a titer of 32 for fold-change calculations if the screen was negative. We defined a confirmed case as a seroconversion with a >4-fold increase in titer from the acute to convalescent sample, in which the convalescent titer was >128. If a participant had a seroconversion to both SFG and TG rickettsiae, we determined the group designation by the higher convalescent titer and then the higher acute titer. We excluded rickettsial infections with an alternative nonrickettsial positive microbiological result on the basis of our protocolized testing (malaria antigen positive, tuberculosis PCR positive or urine lipoarabinomannan positive, positive blood cultures, or positive whole blood Biofire panel) from this analysis to decrease the risk for misclassification (Appendix Figure 1).

For RT-PCR, we extracted RNA (targeting mRNA and rRNA to optimize sensitivity) (10) from 200 µL acute serum and 200 µL acute whole blood by using QIAamp RNA Mini Kits (QIAGEN). We conducted RT-PCR by using previously published methods (18) targeting SFG rickettsia sca0 and TG rickettsia 17-kDa outer membrane lipoprotein mRNA. We only considered samples positive if in duplicate. To develop the 16S rRNA primers and probes, we aligned genes across 41 rickettsial species and 17 clinically relevant nonrickettsial bacterial species downloaded from GenBank by using the MEGAX platform (19). We identified a conserved region that was disparate from nonrickettsial species, including the forward primer 5′-gcgggtaatgccgggaactataag-3′, reverse primer 5′-ccgaactgagatgtcttttaggg-3′, and probe 5′-/56-FAM/gccggagga/zen/aggtggggacgacgtc/3IABkFQ/-3′. To determine primer species specificity or exclusivity compared with off-target organisms, we compared detection and SYBR Green melting curves by using 14 rickettsial and 4 nonrickettsial DNA controls. We used a 175-bp target sequence cloned into the pCR2.1 vector as a plasmid for quantification with a quantitative PCR master mix (Bio-Rad Laboratories). We identified a cutoff for rRNA RT-PCR by using the lowest threshold with >90% detection among serially diluted whole blood samples and in serum samples from healthy donors spiked with cell-free R. parkeri. To determine RT-PCR clinical sensitivity and specificity, we used IFA seroconversions as the index comparator. We prioritized samples with limited volume for IFA and then 16S RT-PCR, sca0, and 17-kDa RT-PCR. The rRNA RT-PCR performance calculations were limited to those with sample availability for both complete paired IFA and RT-PCR testing for serum and whole blood (n = 172 participants).

Our study found that rickettsioses were a common cause of hospitalized illness across multiple sites in Uganda, consistent with a high observed seroprevalence. Rickettsial illness manifested similarly to nonmalarial illness on the basis of well-described clinical parameters. Clinical factors (including platelet count) were neither sensitive nor specific for identifying rickettsial infections. Rashes were infrequently reported or seen on physical examination, and no eschars were observed. Specific rickettsial treatments were uncommonly used. rRNA RT-PCR sensitivity in serum samples was considerably higher than targeting rickettsial mRNAs, although the use of rRNA RT-PCR has the potential benefit as a diagnostic strategy for identifying acute cases that otherwise would not be treated. Our findings highlight the need for continued diagnostic development. Clinicians should also have a high level of suspicion and low threshold for empiric doxycycline use in this region for nonmalarial illness among hospitalized adults.

Among prior cohorts in Sub-Saharan Africa, clinical descriptions of rickettsial infections have largely been limited to returning travelers, and the course of hospitalized illness is not prospectively well characterized. The high rates of seroprevalence observed and common identification of seroconversions indicate that rickettsial infections are circulating widely in this region. A high number of rickettsial infections is supported by a recent study that identified SFG rickettsia in ticks throughout Uganda (20). We found rickettsial infections were common (8%) among patients with acute febrile illness or sepsis manifestations. This result is similar to an estimate from 2012–2014 among pediatric and adult febrile participants in Tanzania in which 8.9% had SFG rickettsia seroconversions (21). Our findings greatly strengthen the evidence that rickettsial infections frequently cause hospitalization in this region.

Novel acute diagnostics are needed to identify rickettsioses. Diagnostic delays for rickettsial infections often result in prolonged hospital stays and increased death because rickettsial infections are not susceptible to standard empiric antimicrobial drugs used in many worldwide settings (2,22,23). The current diagnostic standard, indirect IFA, relies on referral acute and convalescent serum sample confirmation testing, which is only available 3–6 weeks after symptom onset and has performance limitations (9). Our observations of rRNA RT-PCR–positive IFA-negative participants highlighted challenges with using a 4-fold IFA titer increase. In addition to logistical barriers, IFA interpretation could be affected by blood collection timing (24), early receipt of antimicrobial drugs (25), and immunocompromising conditions (26). Characteristics of rickettsial infection were not reliable as clues for initiating empiric treatment, further demonstrating the diagnostic gap and need to consider empiric treatment without fully relying on clinical laboratory results in the absence of a sensitive rapid diagnostic test.

We demonstrated potential improvement in sensitivity by using RT-PCR rather than PCR and by using highly abundant and stable rickettsial rRNA targets rather than membrane protein gene DNA or mRNA targets. Molecular amplification typically has low sensitivity (≈7%–43%) for rickettsioses because of low organism burden; sensitivity can be improved by extracting nucleic acid from the buffy coat, which requires additional processing (10,18). By using multiple singleplex RT-PCR primers (targeting membrane protein gene DNA or mRNA) we were able to detect cases not identified with DNA PCR. However, sensitivity remains limited with either RT-PCR or PCR by using traditional targets present in either low numbers or that are labile. We anticipated whole blood would contain more rickettsial organisms and improve sensitivity compared with serum across PCR targets. Instead, we found that whole blood was less sensitive than serum, potentially because of inhibitors in whole blood (27). rRNA RT-PCR improved sensitivity of PCR over membrane protein gene DNA and mRNA targets and demonstrated promise as an acute detection diagnostic modality that detected rickettsial cases missed by DNA, but mRNA detection was not higher than anticipated when compared with previous studies using DNA (18).

Our results are supported by prior efforts that showed improved performance by using 23S rRNA targets (14,16). The stability and abundance of protein-associated rickettsial rRNA targets are promising for use in remote settings where sample handling and storage can be more vulnerable to unstable conditions. However, rRNA still lacks sensitivity compared with paired serologic testing. Scalable methods to improve specificity and potentially sensitivity such as digital droplet PCR or concentrating target templates in each reaction are needed before broad clinical use (18). However, because RT-PCR is now widely available, potentially with increased capacity after the COVID-19 pandemic, and doxycycline is generally well tolerated, laboratory-validated rRNA RT-PCR could be used to initiate standard of care treatment with doxycycline and then serologic confirmation. Rickettsial rRNA genes are abundant stable targets that could be leveraged for furthering clinical rickettsial diagnostics, but implementation and validation studies are needed at clinical sites. Although performance could be optimized, rickettsial rRNA RT-PCR might be the best option for clinical use in highly endemic regions in the absence of a reliable acute diagnostic alternative.

The first limitation of our study is that it relied on availability of convalescent-phase serologic testing to identify approximately half of the infected patients. Therefore, the severity of each group is not anticipated to be representative of the population because participants lost to follow-up or who had died before 28 days were not included. Second, in emphasizing sample availability for multimodal case identification, our study was not designed to determine differences in deaths or duration of hospital stay. Third, sample collection, handling and storing conditions at resource-limited clinical sites might have decreased PCR sensitivity. rRNA RT-PCR performance could be improved with direct testing after collection, although rRNA is likely less vulnerable to degradation than mRNA. Last, conserved rRNA targets afford broad detection of rickettsioses but are intrinsically less phylogenetically or taxonomically informative, prohibiting the possibility for species level identification of SFG rickettsiae.

In conclusion, we found rickettsial infections to be common across 2 severe infectious illness cohorts in Uganda. Our case–control study identified that commonly suggested clinical factors for identifying rickettsial infections in sub-Saharan Africa are nonspecific or are generally absent. rRNA RT-PCR improved sensitivity over previously used membrane protein gene DNA PCR and mRNA RT-PCR and requires further clinical validation to ensure specificity when using conserved stable rRNA targets. Until acute diagnostics are widely available for rickettsial infections, empiric doxycycline should be considered for nonmalarial fever of unknown cause in this region.