Febrile patients admitted to remote hospitals in Northeastern Kenya: seroprevalence, risk factors and a clinical prediction tool for Q-Fever

Study area

The study was conducted at Garissa Provincial Hospital (GPH) and Wajir District Hospital (WDH) of Northeastern province of Kenya (Fig. 1). GPH has 224 beds and serves an estimated population of 623,060 individuals, whereas WDH has 120 beds serving about 661,941individuals. These health facilities are also the main referral hospitals of the Garissa and Wajir counties respectively. The province has an arid and semi-arid lands (ASAL) climate and is almost exclusively inhabited by the Somali ethnic groups who are predominantly nomadic pastoralists. Livestock keeping is the main economic activity in the region. The common diet for the communities in the province is meat and milk [15, 30].

Drinking of raw milk is common in this population due to deep-rooted cultural traditions. It is believed that boiling milk reduces the nutritional value of milk. Milk is also mixed from several animals in common containers. From this milk, fermented products (local name ‘susac’) are prepared. Susac would then be sold at the nearest open market places and nomadic camp sites [15, 31]. This province was selected because public health support and disease surveillance has been extremely limited and outbreaks of re-emerging diseases have been reported in the province before [32–34].

Study participants and procedures

The hospital-based study was conducted between June 2014 and January 2015. Acute febrile patients presenting to the outpatient departments of the two hospitals were systematically assessed for enrollment. The inclusion criteria were: a clinical history of acute febrile illness (AFI) characterized by fever (higher than or equal to 38 °C) and at least one of the following clinical symptoms: headache, chills, myalgia, arthralgia, general body malaise, and acute lower respiratory illness (ALRI). ALRI was defined as new onset of cough, difficulties when breathing or chest pain and fever [17]. A study clinician then collected demographic information, obtained clinical history and performed physical examination of the enrolled participants. The clinical diagnosis made by the attending hospital clinician was also recorded. Exclusion criteria were: patients younger than 18 years not accompanied by a guardian or parents, patients who had an already established diagnosis and those who were severely sick and were considered not able to provide informed consent. Patients who declined to participate were also excluded.

Sample and data collection

Demographic information and data on putative social-economic and dietary characteristics of the patients were collected using pre-tested questionnaires. Risk factors were assessed in form of a mix of closed- and open-ended questions including the following data: (a) living in a livestock keeping household/nomad camp or involvement in animal husbandry activities for at least 2 days a week (hereafter referred to as ‘regular contact with animals’), (b) animal species involved, and (c) the type of contact with animals or their products i.e. help in animal parturition, contact with aborted materials, grazing animals, livestock trade, handling raw milk, and animal slaughter. Participants were also asked if they shared their house with animals and if they were involved in manure preparation. The food variables assessed were: (i) consumption of raw milk, (ii) locally fermented milk products (susac), (iii) drinking of animal blood and (iv) consumption of uncooked meat.

Following cleansing of the skin with isopropyl alcohol and povidone iodine, five milliliters of venous blood were collected aseptically into empty 10 ml serum BD vacutainer tubes® (Becton, Dickson and Company, USA). The blood samples were stored at room temperature for 45 min to allow blood clotting. The samples were then centrifuged at 2500 rpm for 5 min and serum was separated. Sera were aliquoted into cryo-vials before shipping to Friedrich-Loeffler-Institut (FLI) for Q fever laboratory analysis.

Serological analysis

Sera were analyzed for IgG antibodies against phase I and phase II antigens using serion ELISA classic test kits (Virion\Serion, Würzburg, Germany) and evaluated according to the manufacturers’ recommendations. For phase I, samples with an optical density (OD) >10 % above OD of the cut-off sera were scored as positive, those with ODs >10 % below OD of the cut-off sere were considered as negative, in between samples were denoted as borderline. For phase II, the cut off value was calculated on the basis of the standard curve corrected by the mean of the extinction of the standard serum according to the manufacturers’ instructions. A result of <20 U/ml was regarded as negative, 20–30 U/ml borderline, and >30 U/ml as positive.

The seropositive samples and those with inconclusive results were confirmed with a commercial IFA for detection of IgG antibodies against C. burnetii phase I and phase II (Fuller Laboratories, USA). The assay titers were started at 1:16 and results interpreted according to the manufacturers’ instructions. Samples with C. burnetii phase II antibodies titers of >1:128 were scored as acute Q fever positive, while those with phase I antibodies titers of ≥1:800 were regarded as positive for chronic Q fever. The samples that were found to be equivocal in the repeat tests were re-tested at the Q fever consulting laboratory Baden-Wurttemberg, State Health Office, Germany.

DNA extraction from samples

DNA isolation was done using the High Pure PCR Template preparation commercial Kit™ (Roche Diagnostics, Germany) according to the manufacturer’s instructions. Purity and concentration of DNA was tested using a Nano Drop ND-1000 UV-vis spectrophotometer (Nano Drop Technologies, Wilmington, USA), and DNA samples were stored at −20 °C until analyzed.

Quantitative Real-time PCR assay (qPCR)

Detection of C. burnetii DNA was performed with a TaqMan based qPCR assay targeting the repetitive element IS1111 as described by Klee et al. [35] using a Stratagene Mx3000P v 4.01Thermocycler (Agilent Technologies, Santa Clara, USA). No template control (NTC) and tenfold serial dilution of cloned IS1111 gene plasmid fragments ranging from 1 × 10⁰ to 1 × 10⁵ plasmid copy numbers were used for Coxiella DNA quantification and sensitivity control of the assay. Cycling conditions were as follows: one pretreatment cycle at 50 °C for 2 min, initial denaturation at 95 °C for 10 min, followed by 50 cycles at 95 °C for denaturation for 15 s, and 30 s for annealing and elongation at 60 °C. No internal amplification control (IAC) was used in the procedure to ensure high sensitivity. Cycle threshold (Ct) values ≤ 36 cycles were interpreted as positive. The threshold was calculated automatically by the instruments’ software based on six 10 fold serial dilutions of C. burnetii DNA in negative human sera assessed simultaneously in a single run.

Case definitions

Acute Q fever case was defined as a patient with compatible clinical evidence criteria and laboratory supportive results for acute Q fever illness, reflected by elevated IgG phase I and phase II antibodies by ELISA and confirmed by C. burnetii phase II antibodies titers of >1:128 by IFA assay or detection of Coxiella DNA by qPCR. Chronic case was defined as those with elevated IgG phase I antibodies by ELISA and phase I antibodies titers of ≥1:800 by IFA assay. The case definition criteria was based on a combination of recommendations of the kits manufacturer and as proposed by EFSA [2], CDC [7], and previous authors [8, 9, 36]. In all cases, Salmonella spp, Brucella spp, and Plasmodium spp were excluded. For statistical analysis, seropositivity was defined as any participant who met the criteria for acute or chronic Q fever.

Data management and statistical analysis

Questionnaire results and serological data were entered into Microsoft excel 2010 spreadsheet and exported to SPSS Statistics software® (Armonk, IBM Corp, USA. v.20) and verified against the paper questionnaires and laboratory sheets for consistency and completeness. The Pearson’s χ²test or Fisher’s exact test was used to determine differences in Q fever seroprevalence among demographic groups. Statistical significance was set at P value < 0.05.

Multivariate logistic regression models were built to assess clinical features, and plausible socio-demographic, economic and dietary characteristics of the patients associated with Q fever seropositivity using stepwise backward analysis procedure. Briefly, univariate analysis was performed for all covariates in each model and pair-wise collinearity assessed using Spearman rank correlation. A pair of the variables was considered collinear if the correlation coefficient was equal or greater than 0.8. Multi collinearity was then assessed for all potential variables by estimating the variance inflation factor (cut-off 5) and tolerance (cut-off 0.2) levels using collinearity diagnostics function (IBM Corp, USA. v.20). The variables were then droped or aggregated as appropriate.

The candidate covariates with p < 0.2 using (Wald test) were fitted into the multivariable model and those exhibiting the highest p-values (Wald test) were removed one at a time from the model until all the retained variables had p < 0.05. The eliminated variables were re-screened for confounding and were retained if their inclusion caused 20 % or more change to the coefficients of one or more of the retained variables. Age was considered to be a biologically important variable and thus was kept in the final model. Finally, both the clinical and putative socio-economic and dietary characteristics from the two sub models were combined in a multivariate analysis to identify the independent risk determinants for Q fever seropositivity. The Hosmer-Lemeshow test was used for assessing final models’ goodness of fit (GOF). The area under the ROC curve was also generated to assess the predictive ability of the final best model.

To derive and evaluate a predictive clinical model best-fit for identification of a subgroup of patients with high likelihood of acute Q fever, the patients were randomly assigned to derivation set and validation set using random number generated using the SPSS random number generator function (IBM Corp, USA. v.20). The derivation set used to fit the model composed of 65 % of the sample (n = 707) while the validation set used to validate the model was made up of 35 % of the sample (n = 360) [37]. A clinical prediction score “Q fever score” was then generated using the β coefficients of the final best fit model and the discriminatory power of this score evaluated by the area under the receiver operating characteristics (ROC) curve at the 95 % confidence interval (CI). Then a cut-off value to estimate the diagnostic sensitivity and specificity in the validation set was selected [38].

Study population

Seroprevalence

A total of 204 (19.1 %, 95 % CI: 16.8–21.6) participants were found to be positive for Q fever based on a parallel interpretation of the three tests applied. Of these, 181 (17.0 %, 95%CI: 14.8–19.4) had anti-Coxiella phase II IgG antibodies while 175 (16.4 %, 95%CI: 14.2–18.8) had IgG antibodies against phase I antigens. One hundred ninety-four (18.2 %, 95 % CI: 15.9–20.7 %) patients were positive for IgG antibodies against either phase II or phase I antigens, whilst 162 (15.2 %, 95 % CI: 13.2–17.5) patients were found to be positive for anti-Coxiella IgG antibodies against both phase I and phase II antigens (Table 1). Thirteen and eighteen participants tested positive for only IgG phase I and IgG phase II antigens, respectively. One patient remained positive for IgG phase II and borderline for phase I in the repeat tests. Based on the study case definition criteria, acute Q fever was established in a total of 163 (15.3 %, 95 % CI: 13.2–17.6) patients whereas none of the patients had IgG antibody titers against phase I supportive of chronic Q fever infection.

Molecular detection of Q fever

Quantitative real-time PCR (qPCR) demonstrated C. burnetii specific DNA in 2.2 % (10/448) of the examined patients, of which all were serologically negative for Q fever. The tested sera were from patients whose onset of the illness was ≤ 2 weeks and had no antibiotic therapy. The PCR of whole blood or serum is mainly positive within the first two weeks following symptom onset, and becomes negative as the antibody response develops. Detectable seroconversion typically occurs one to three weeks after symptoms appear [39].

Q fever was not clinically suspected by the treating clinicians in any of 173 (16.2 %) positive patients, instead the working diagnosis were mainly presumptive typhoid fever (45.1 %), malaria (6.9 %), pneumonia oracute respiratory infections (37.6 %) and others or fever of unknown origin (FUO) (10.4 %). There was no significant difference in infection found between males and females (P = 0.736). Higher seroprevalence (16.7 %, 95 % CI: 14.4–19.3) was found in adults (>18 years) when compared to those aged below (13.0 %, 95 % CI: 8.0–20.2), but the difference was not statistically significant (P = 0.283). Somalis were more often seropositive when compared with other ethnic groups (Somali, 17.1 %, 95%CI: 14.8–19.7 and Non-Somali, 7.7 %, 95%CI: 3.6–15.4, P = 0.004, Table 1). Among 885 participants who reported regular contact with animals, 160 (18.1 %, 95%CI: 15.7–20.8) tested positive for Q fever whereas only 13 (7.1 %, 95 %: CI: 4.10–12.2) who reported no contact with animals were positive (P < 0.001). Most cases of Q fever occurred in full time herders 134 (19.8 %, 95%CI: 16.9–23.1) and livestock trader 15 (25.4 %, 95%CI: 15.4–38.7) when compared to government workers, general merchandise businesses and other occupations (P < 0.001). Patients from Wajir county were slightly more seropositive 90 (16.8 %, 95%CI: 13.8–20.3) than those from Garissa county 83 (15.6 %, 95 %: CI: 12.7–19.1) but the difference was not statistically significant (P = 0.607). Finally, patients without knowledge of zoonosis 115 (18.6 %, 95%CI: 15.6–21.9, P = 0.012) and those without any form of education 130 (19.7 %, 16.8–22.9, P < 0.001) were more often seropositive (Table 1).

Clinical predictors of acute Q fever diagnosis and ‘Q fever’ score

Univariate analysis of the signs and symptoms revealed that general body malaise and fatigue, ALRI, constipation, abdominal pain, diarrhoea, and fever onset (>2 weeks) were possibly associated with Q fever seropositivity (P < 0.20; Table 3).

In the final clinical logit model, adjusted for possible confounding variables, ALRI (adjusted Odds ration [aOR]: 2.68, 95%CI: 1.65–4.36), fever onset (>2 weeks) (aOR: 37.59, 95%CI: 17.83–79.27) and abdominal pain (aOR 2.19, 95%CI: 1.02–4.72) were positive predictors of Q fever infection. The analysis revealed that diarrhoea (aOR: 0.34, 95%CI: 0.12–0.96) had predictive value for a negative acute Q fever outcome (P < 0.05; Table 4). Using the four variables, predictive score was generated and the scale simplified by assigning values to the nearest one decimal scale (fever onset >2 weeks: +3.6, ALRI: +1.0, abdominal pain: + 0.8 and diarrhoea: −1.1, Table 4). The discriminatory power of the fitted ROC curve area was 0.883 (95%CI: 0.851–0.915, P < 0.001, Fig. 2a). An optimal cut-off value of 2.90 (sensitivity 93.1 % and specificity 76.1 %) was selected (Fig. 2b). Only a history of fever (>14 days) singly was a good predictor of positive diagnosis (sensitivity 90.5 % and specificity 76.8 %). The prediction remained strong, if diarrhoea was absent or when at least one of positive predictors was also present. However, none of other clinical predictors (ALRI or abdominal pains) alone or in combination had significant predictive value for positive acute Q fever outcome. Thus, pre-admission fever onset (>14 days) in association with ALRI and abdominal pain and no diarrhoea was the best predictor algorithm for acute Q fever outcome. The prediction rule, reliably identified febrile patients with a score >2.90 as 12.7 times more likely to have a positive acute Q fever test outcome (OR = 12.7; 95 % CI: 4.89–32.72) than the patients with a Q fever score up to 2.90.

Variable	Crude OR (95%CI)	aOR (95%CI)	β Coefficient	P- value
ALRI
No	Ref	Ref		-
Yes	3.61 (2.39–5.44)	2.68 (1.65-4.36)	0.986	<0.001
Abdominal pain
No	Ref	Ref		-
Yes	1.57 (1.01–2.42)	2.19 (1.02–4.72)	0.788	0.004
Diarrhoea
No	Ref	Ref		-
Yes	0.60 (0.28–1.29)	0.34 (0.12–0.96)	−1.075	0.042
Fever onset (>14 days)
No	Ref	Ref		-
Yes	41.52 (19.78–87.22)	37.59 (17.83–79.27)	3.627	<0.001
H-L test				0.567
ROC (AUC)	0.883 (0.851–0.915)			<0.001
Q fever score	Q fever score (rounded) = 1.0x(ALRI) + 3.6x(Symptoms onset (>14 days) + 0.8x (Abdominal pains) -1.1x(Diarrhoea).

ALRI acute lower respiratory infection, Ref referent category, aOR adjusted odds ratio, CI 95 % confidence interval, H-L test Hosmer-Lemeshow goodness- of-fit test, ROC receiver operating characteristics, AUC area under the curve

Risk factor analysis for acute Q fever

Univariate analyses of demographic, social-economic and dietary characteristics showed no significant associations with seropositivity between gender, county of residence of the patient or across different age strata, and occupations except herders (OR; 2.96, 95 % CI: 1.15–8.34) and livestock traders (OR; 4.09, 95 % CI: 1.26–13.27) who had significantly higher odds for seropositivity. Having secondary education or above (OR; 0.39, 95 % CI: 0.17–0.81) and knowledge about zoonosis (OR; 0.78 CI: 0.36–0.99) were associated with decreased risk of Q fever seropositivity. Odds of Q fever seropositivity were significantly higher for patients of Somali ethnicity (OR; 1.92, 95 % CI: 1.29–3.76).

Regular contact with cattle (OR; 2.65, 95 % CI: 1.89–3.72) or goats (OR; 4.82, 95 % CI: 2.89–8.26) were independently associated with Q fever seropositivity. The risk for Coxiella infection was higher for those who reported the following: contact with aborted materials or assist in animal births (OR; 1.43, 95 % CI: 0.94–2.19), handling raw milk (OR; 1.55, 95 % CI: 1.10–2.18), grazing animals (OR; 1.98, 95 % CI: 1.34–2.92), and slaughtering of animals (OR; 3.44, 95 % CI: 2.46–6.67). Q fever seropositivity was significantly associated with consumption of raw cattle milk (OR; 6.43, 95 % CI: 4.52–9.14), raw goat milk (OR; 1.77, 95 % CI: 1.20–2.61) and locally fermented milk products (susac) (OR; 3.81, 95 % CI: 2.73–5.36) (Table 2).

In the final combined multivariate analyses, significant association persisted for pre-admission fever (>14 days) (aOR; 36.35, 95 % CI: 19.84–69.49) and ALRI (aOR; 2.41, 95 % CI: 1.53–6.36), while abdominal pain (aOR; 1.95, 95 % CI: 1.28–3.58) association with seroposivity was strengthened.

Regular contact with cattle (aOR: 2.09, 95 % CI: 1.73–5.98), or goats (aOR: 3.74, 95 % CI: 2.52–9.40) and frequent slaughter of animals (aOR: 1.78, 95 % CI: 1.09–2.91) remained with significantly higher odds of seropositivity. The dietary exposures significantly associated with seropositivity in the final combined model included: consumption of raw cattle milk (aOR: 2.49, 95%CI: 1.48–4.21) and locally fermented milk products (aOR: 1.66, 95%CI: 1.19–4.37).

Limitations

A potential limitation in our findings is that our study was hospital based where the population under investigation was febrile patients seeking treatment. Therefore, the afebrile cases or those that were unable to present to the hospitals were inevitably not captured by our sampling strategy. Again, the patient samples were generated in absence of a robust probabilistic sampling method. This has the effect of limiting the generalization of the results. A population-based survey or follow up of patients to obtain a convalescent sera sample was not feasible due to the nomadic lifestyle of the patients, the ongoing inter-clans conflicts and militia activities in the region. However previous studies have reported adequate performance of the used commercial serion ELISA classic test kits in detection of phase II specific IgG and IgM [36, 62] respectively, in diagnosis of acute Q fever. The novelty of the present study was that the ELISA positive sera were further confirmed with the reference diagnostic method (IFA) and the patients whose fever onset was ≤ 2 weeks were further screened for Q fever using a quantitative real-time PCR. Nevertheless, the possibility of recent infection cases, false positive or chronic cases being considered as acute Q fever cannot be completely rule out.

The utility of our prediction score should be interpreted in the context of potential limitation. The cross-sectional nature of the study did not allow more comprehensive evaluation and follow up of each patient and the fact that the prediction rule was derived by logistic regression modeling which can be prone to residual confounding effects. Another potential limitation is related to the heterogeneous clinical presentation of Q fever in patients and the fact that the score was validated in patients from the same region where it was developed. Though deliberate efforts were made to exclude the commonly reported febrile etiologies in the region including typhoid fever, malaria and brucellosis, unfortunately due to the nature of the present study and scarcity of appropriate laboratory and radiological support, a comprehensive differential diagnosis for other febrile etiologies among our patients was not feasible. These factors may have affected the discriminatory index of the present score. Nevertheless, our study was not designed to give a definite clinical algorithm but one to help clinicians in distinguishing febrile cases probably due to Q fever from febrile episodes not due to Q fever and aid in making clinical empiric therapeutic decisions. Therefore, validation of the present prediction score in well-designed studies in diverse independent cohorts is highly warranted.

General recommendations

A one health approach for surveillance of emerging and remerging infectious diseases should be encouraged in Kenya to support design and implementation of rational control strategies for febrile diseases. Education programmes and appropriate preventive interventions targeting the significant risk factors needs to be designed for the communities at high risk of exposure. Increased clinician and laboratory personnel awareness and access to laboratory testing capacity is needed to enable timely detection and early treatment to prevent severe sequelae.