Intensive care support and clinical outcomes of patients with Ebola virus disease (EVD) in West Africa

This retrospective cohort study includes all consecutive patients affected by EVD hospitalized in two ETCs in Sierra Leone, from 19/09/2014 to 12/12/2014 in the Lakka Holding and Treatment Centre and from 12/12/2014 to 28/05/2015 in the Goderich ETC-ICU. Both ETCs were located near Freetown, Sierra Leone. More detailed information on the staff, facility, and operational procedures are provided in the Supplementary Material, S1 and S2. The main characteristics and differences are reported in Table 1.

The plasma concentration of Ebola virus (EBOV) RNA (viral load) was measured using a real-time reverse transcriptase PCR (RT-PCR) assay (RealStar Filovirus Screen. RT-PCR 1.0 kit, Altona Diagnostics), with a limit of detection of 10³ cp/ml of EBOV RNA. Viral RNA quantification was based on a standard reference curve provided by the kit manufacturer, spanning up to 10⁹ cp/ml of EBOV RNA. For the sake of simplicity, further analyses were performed using the base 10 logarithm of the viral load (LVL) measured in cp/ml.

At the start of the epidemic, before implementing its own virology laboratory, EMERGENCY collaborated with the WHO South Africa Ebola laboratory, which used the same diagnostic test.

Patients reaching the facility with suspected EVD were admitted, under observation, to the Holding and Treatment Centre, and their symptoms treated while awaiting the RT-PCR test results. Only patients with laboratory-confirmed EVD from the holding centre, or those with EVD transferred from other facilities, were admitted for treatment.

Demographic information including age, gender, symptoms, and possible time of Ebola exposure were collected on standardised clinical case report forms. LVL was recorded from the first diagnostic blood sample at admission. Other risk factors at admission, including markers of organ damage (renal, liver, coagulation etc.), were not systematically collected in the early months and cannot therefore be included in the analysis.

A survivor is defined as a patient discharged alive. Discharge of patients from the ETCs occurred after two negative RT-PCR blood tests performed at least 24 h apart.

Fluids and electrolytes were administered by mouth or intravenously according symptoms, diuresis and fluid balance; all patients were treated for symptoms as nausea, anxiety, and pain. Antibiotics were administered from admission and antimalaria treatment was discontinued only if negative tests became available.

Only a few doses of compassionate use investigational antiviral treatments were available. Two patients received ZMAb and towards the end of the outbreak, Goderich ETC took part in the Prevail II trial [8], randomizing five patients to standard care plus ZMapp (two patients) versus standard care alone (three patients). Based on the available preclinical literature suggesting in vitro and in vivo efficacy [24], a randomised clinical trial protocol was developed to investigate the efficacy of treatment of EVD with amiodarone. The trial was registered at ClinicalTrials.gov (NCT02307591) (see Supplementary material S4 for details).

While awaiting clinical trial initiation, and after approval by two Italian Ethics Committees, given the pre-existing high mortality rate and lack of known effective therapies, amiodarone was administered on a compassionate basis to 100 patients. Compassionate use was stopped between December 2014 and February 2015 by request of the Sierra Leonean Government, pending the opinion of the specific Ethics Committee (definitive approval granted April 23, 2015). Further information on the treatment with amiodarone is reported in the Supplementary material S4.

Descriptive analyses were presented as frequencies, proportions, means and standard deviations, medians and interquartile ranges (IQR), where appropriate to characterize patients and their clinical outcomes. Proportions were compared using the χ² test or Fisher Test, whereas comparisons between continuous variables were expressed through the t test or Wilcoxon rank-sum test, as appropriate. The number of hospital-free days within 28 days was calculated to compare the burden of EVD in terms of hospitalization in the two ETCs, avoiding bias due to mortality rate differences [25]. During the 28-day follow-up period most, but not all, EVD patients either died or recovered. The number of hospital-free days for a patient with x days of hospitalisation was 28 − x, if the patient recovered within 28 days, and 0 if the patient recovered after 28 days or died. Hence, the higher the value of this indicator the lower the burden of hospitalization (as the patient recovered earlier), while deceased patients were considered equivalent to alive patients with a long recovery time.

Multivariable logistic regression models were used to determine independent factors associated with EVD patient mortality. ETC was forced in the model to explore the secondary endpoint of association of intensive care availability on mortality. To test any relationship between the logit and the LVL, the former was modelled as a continuous piecewise linear function of the latter. The Akaike Information Criterion (AIC) [26] was minimised with respect to the cut-off values of LVL where the slope of the piecewise linear function changed. Interactions between ETC and the other variables were only included when they significantly reduced the AIC. The interaction between ETC and the above-described transformation of the viral load was also tested in this step. This allowed different piecewise linear transformations (parametrised by the number and values of the cut-offs) of the LVL for the two ETCs. Missing records were excluded from the analysis.

The odds ratio (OR) of the two ETCs, i.e., the exponential of the difference d between the logit of the two ETCs (d = logit_Lakka – logit_Goderich), was plotted as a function of viral load, with the bootstrap-based confidence band. Hence, mortality in Lakka was higher than in Goderich when OR was larger than one. The OR was statistically significant for those viral load values for which the confidence band was above the line OR = 1. To identify the range of clinically relevant phenomena, two confidence bands were constructed at the 95 and 80% levels, respectively.

The adjusted probability of death was then plotted for each ETC with the respective confidence bands to determine how mortality varied as a function of viral load in the two ETCs. All variables except LVL and ETC were chosen to be equal to their mean values. The width of the confidence bands for Lakka and Goderich was estimated by Payton et al. [27] approximate method. This procedure is designed to evaluate, with a given significance level, whether two confidence intervals overlap (Supporting information, S5).

The role played by viral load at admission as a predictor of mortality within 20 days of hospitalisation, was also described using Kaplan–Meier curves in strata of LVL, overall and for each ETC (Fig. 2). The cut-offs separating the strata were determined as the LVL values at which mortality in one ETC becomes higher or lower than in the other ETC at the 80% confidence level (as described in the previous paragraph). Graphically, these values correspond to the points where the 80% confidence band of the logit difference d crosses the horizontal axis d =0. Differences between Kaplan–Meier curves in these strata were assessed by the log-rank test.

More details on the statistical elaboration are reported in Supplementary materials, S5.

Goderich ETC was built during the outbreak, in the framework of a project to build seven Ebola treatment centres funded by DFID (Department for International Development—GOV.UK). In accordance with treatment criteria and based on the previous 3 months’ experience, the Centre was co-designed by EMERGENCY’s technical/engineering staff. Fitting it out for intensive care delivery resulted in additional costs, borne by the NGO, in the amount of approximately €450,000. The Istituto di Malattie Infettive “Lazzaro Spallanzani”, which ran the virological laboratory, was supported by a “Ricerca Corrente IRCCS” grant; Ministry of Foreign Affairs, Italy (See Supplementary material S2 for more details). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data were analysed on 229 patients with a laboratory-confirmed diagnosis of EVD, admitted to the two ETCs (123 at the ETC in Lakka, 106 at the ETC-ICU in Goderich). All patients with confirmed EVD received supportive treatment and were followed up to the endpoint (recovery or death). The study flow chart is shown in the supplementary material (Fig S3.1).

No major differences emerged between ETCs for admission viral load, or time from symptom onset to ETC-admission (Table 2).

The mean number of hospital-free days (of 28) days of admission was significantly higher in the Goderich ETC-ICU (7.7 vs 5.5 days; p = 0.03, Table 3).

The overall crude case fatality rate (CFR) was 54.1% (Table 4). No statistical differences were observed between patients who survived and who died of EVD in terms of gender, age and time from symptom onset to hospital admission. CFR differed substantially according viral load at admission, ranging from approximately 30% for patients with LVL < 7.5 (low) to over 80% for patients with LVL ≥ 8.5 (high) (p < 0.0001). Crude analysis showed a non-statistically significant difference in CFR between Lakka-ETC and Goderich-ETC-ICU patient admissions (57.7 vs 50.0%; p = 0.19) (Table 4). Multivariable logistic regression analysis showed that age and viral load were independently associated with an increase in EVD mortality. The best model (according to the lowest AIC) for predicting the probability of death included age, LVL, ETC, and the interaction between LVL and ETC. As known, an interaction involving a continuous variable (like LVL) is difficult to represent. Hence, in Fig. 1 we plotted, as a function of LVL, both the OR for mortality of Lakka vs Goderich (left panel) and the probability of death at each ETC (right panel). In the left panel, the 95% confidence band of the OR (light grey) never exceeds one, while the 80% confidence band (dark grey) exceeds one for medium–high viral loads (LVL between 7.5 and 8.5).

Figure 2 shows the importance of the viral load at admission for the outcome of the overall population as Kaplan–Meier survival-curve for patients with low, intermediate and high viral loads at admission (panel a). The same analysis split in patient population of Lakka (panel b) and Goderich (panel c) shows how the intermediate line, corresponding to patients with LVL 7.5–8.5, is different in the two settings.