Introduction

Sepsis is a dysregulated host response to infection that results in life-threatening organ dysfunction [1]. Based on data from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017, an estimated 48.9 million people worldwide experienced sepsis in 2017, with 11.0 million deaths [2]. This burden is particularly high in low-income countries (LICs), with an estimated 16.7 million cases of sepsis occurring annually in sub-Saharan Africa [2]. A meta-analysis of studies in sub-Saharan Africa demonstrated a sepsis mortality rate of 19% and a severe sepsis mortality rate of 39% [3]. In Rwanda, the mortality in two intensive care units was shown to be 64.4% for sepsis and 82.1% for septic shock [4].

Although these data demonstrate a high incidence and mortality from sepsis in LICs, few studies on sepsis presentation, management, outcomes, and predictors of mortality in LICs have been reported. In a study assessing critically ill patients with suspected infection in a public tertiary referral center in Kenya, the majority (65.1%) were suffering from respiratory infections and the most common comorbidity was diabetes mellitus (27.3%) [5]. In a survey of critical care providers at a large tertiary center in Kenya, respondents indicated the most common sources of infection in sepsis were respiratory and intra-abdominal, the most frequently used antibiotics were ceftriaxone and metronidazole, and 43% of respondents ordered blood cultures on suspicion of sepsis regularly [6]. In a private hospital in rural Uganda, a study in 20 adults and 31 children with sepsis found that the most common comorbidity in adults was HIV (30%); greater than 80% of patients received an antibiotic; and half of all patients with an elevated lactate received intravenous fluid resuscitation [7]. In a referral hospital in the capital city of Haiti, the most common sources of infection in patients with sepsis were lung and intra-abdominal [8]. Management at this facility consisted of intravenous fluid resuscitation in 80% of patients with severe sepsis, and 54.6% of patients with severe sepsis received antimicrobials within 24 hours, with the most commonly used antimicrobials being ceftriaxone and chloroquine.

Mortality from sepsis in most of these observational studies at large referral centers ranged from 20.4% to 24.2%, yet mortality in the study at the rural hospital in Uganda was only 3.9% [5–8]. The study in Haiti found encephalopathy, supplemental oxygen therapy, and stool microscopy to be predictors of mortality among septic patients [8]. Another study in Uganda found Glasgow Coma Scale (GCS), Karnofsky Performance Scale (KPS), tachypnea, thrombocytopenia, and leukocytosis to be mortality predictors [9].

The optimal treatment of sepsis in low-income countries is unknown. The only randomized controlled trial of sepsis resuscitation in adults in Africa demonstrated harm with fluid resuscitation similar to that considered standard of care in HICs [10]. This may be related to late presentation, underlying infection types that differ from those in HICs, and lack of access to mechanical ventilation after resuscitation [10–12].

We aimed to evaluate the presentation, management, outcomes, and predictors of mortality from sepsis through a retrospective cohort study of patients admitted to two of the three public tertiary hospitals in Rwanda.

Materials and methods

Results

Demographic and baseline characteristics

A total of 181 charts met inclusion criteria (100 at CHUK and 81 at CHUB) and were reviewed. Nearly three quarters of our sample were female (72.4%), and the median age was 40 (IQR = 29, 59) (Table 1). Nearly half (48.9%) of the subjects resided in a rural district. The majority (89.0%) of patients were transferred from a district hospital or health center. Of the 113 patients transferred from another health center, we had data on length of stay at the outside facility for 32/113 (28.3%). For those, the median length of stay at these facilities was 2.0 days (IQR = 1.0, 5.5).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Demographic and baseline characteristics.

https://doi.org/10.1371/journal.pone.0251321.t001

One hundred eight subjects (59.7%) met criteria for sepsis at time of hospital admission and 73 met inclusion criteria later in the hospital course. The median time from hospital presentation to meeting criteria for sepsis was 0 days (IQR 0, 1.3). Vital signs were recorded consistently with the exception of respiratory rate and temperature (38.1% and 32.6% of values were missing, respectively) (S1 Table). At time of meeting criteria for sepsis, median heart rate was 112 beats per minute, respiratory rate was 24 breaths per minute, systolic blood pressure was 96 mmHg, temperature was 37.1°C, SpO2 was 95%, and median GCS was 15 (IQR 11, 15) (Table 1). Median white blood cell count, platelet count, and creatinine were within normal limits. Known HIV positive patients represented 19.3% of the sample; 6.1% of the sample had known prior hypertension; 6.1% had diabetes mellitus. Other comorbidities (Hepatitis B or C, malignancy, and chronic kidney, liver, heart, or lung disease) were seen in the subjects at rates less than 5%.

The most common source of infection documented at discharge was intra-abdominal (37.0%) followed by pulmonary (32.6%) (Table 1). Of the 95 patients who received a malaria blood smear, 21 (22.1%; 11.6% of the total cohort) were positive for Plasmodium species. Five patients (2.8% of the total cohort) tested positive for Mycobacterium tuberculosis either by acid-fast bacilli smear or nucleic acid amplification test. Of the 55 blood cultures obtained, 9 (16.4%) were positive. Sensitivities were available for four of these samples, and antimicrobial resistance was present in three samples (S2 Table). Antimicrobial resistance in urine cultures (n = 4) was present in all samples (S3 Table).

Diagnostics and management.

Sepsis or septic shock was listed as a diagnosis in the medical record in two-thirds of cases (Table 2). Septic shock criteria were met by 70 (38.7%) subjects. Patients with sepsis or septic shock were most frequently managed in the internal medicine ward. The majority of patients (n = 95, 52.5%) had performance of a malaria blood smear recorded (S4 Table). Blood cultures were drawn in 55 (30.4%) of patients and 36 (19.9%) had a urine culture performed. Twenty-six patients (14.4%) were assessed for tuberculosis by nucleic acid amplification test and 9 (5.0%) were assessed by acid-fast bacilli smear (S4 Table). Ultrasound of any internal organ was performed in 45.3% of patients and 34.8% of patients received a chest radiograph (S4 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Management.

https://doi.org/10.1371/journal.pone.0251321.t002

The vast majority of patients (92.7%) received intravenous fluid resuscitation within 8 hours and 96.8% received fluid resuscitation within 24 hours (Table 2). The median volume of intravenous fluid administered in the first 8 hours after meeting sepsis criteria was 1.0 L (IQR = 0.5, 2.0 L), and in the first 24 hours the median intravenous fluid received was 2.0 L (IQR = 1.0, 4.0 L). Nearly all patients (94.0%) received at least one antimicrobial, and the most frequently administered antimicrobials were ceftriaxone and metronidazole (S5 Table). Antimicrobials with broader spectrums of activity were frequently added later in the hospital course (S2 Table). Exploratory laparotomy was performed on 23.8% of subjects (Table 2).

Vasopressors were administered in 58 patients (32.0%) and the most frequently administered vasopressor was epinephrine followed by dopamine (Table 2). A total of 46.4% of patients with sepsis received mechanical ventilation. The median duration of vasopressors and mechanical ventilation was 0.7 days (IQR 0.3, 2.1) and 2.4 days (IQR 0.8, 7.2), respectively. Glucocorticoids were administered in 7.7% of cases, and 9 patients (5.0%) underwent central venous catheter placement.

Outcomes

Acute kidney injury was the most common complication (19.3%), followed by acute respiratory distress syndrome (9.9%) (Table 3). The median length of stay in the ICU was 2.3 days and the median length of stay in the internal medicine and surgery wards were 9.4 days and 6.0 days, respectively. The overall median length of stay in the referral hospital was 9.5 days. The in-hospital mortality rate for sepsis or septic shock in this study was 51.4%. Mortality for septic shock was 82.9% and for sepsis without septic shock was 31.5%.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Outcomes.

https://doi.org/10.1371/journal.pone.0251321.t003

Predictors of mortality.

On univariate analysis, presentation characteristics associated with in-hospital mortality were male sex, age, province of residence, residing in a rural district, transfer from a district hospital, presentation to CHUB, high respiratory rate when meeting inclusion criteria, lower GCS score when meeting inclusion criteria, lower platelet count, HIV positive status, and intra-abdominal or central nervous system source of infection. There were no significant interactions between assessed covariates. Baseline characteristics at the time of meeting sepsis criteria with statistical significance that predicted mortality on multivariable logistic regression were one breath per minute increase in respiratory rate (adjusted odds ratio (aOR) = 1.10, 95% CI = 1.01–1.21, p-value = 0.037), one point decrease in GCS (aOR = 0.793, 95% CI = 0.621–0.968, p-value = 0.021), and known HIV+ status (aOR = 7.02, 95% CI = 1.15–55.8, p-value = 0.034) (S6 Table). The area under the receiver operating characteristic curve for this model was 0.861 (95% CI: 0.787–0.934). Additional fit statistics include Hosmer-Lemeshow chi-square statistic of 3.0662 (p-value = 0.930), Brier Score = 0.150, and Nagelkerke pseudo R² = 0.513.

Management characteristics on univariate analysis associated with higher mortality included higher volume of fluid resuscitation in the first eight hours after sepsis presentation, management in the intensive care unit, urine output not recorded by nursing staff, receiving metronidazole or cefotaxime, receiving steroids, receiving vasopressor therapy, receiving mechanical ventilation, and undergoing intubation, exploratory laparotomy, or central venous catheter placement (Table 2). No significant interactions were found between assessed covariates. Nearly all (93.1%) patients who received at least one vasopressor died and the vast majority (83.3%) of those who received mechanical ventilation died. Upon controlling for age, sex, and the predictors identified in the presentation characteristics model (illness severity indicators of respiratory rate and GCS, and known HIV+ status) the management factors with statistical significance associated with mortality on multivariable logistic regression were administration of vasopressors (aOR = 7.46, 95% CI = 1.69–40.2, p-value = 0.007) and mechanical ventilation (aOR = 6.65, 95% CI = 1.04–51.6, p-value = 0.046) (S7 Table). The area under the receiver operating characteristic curve for this model was 0.895 (95% CI: 0.829–0.961). Additional fit statistics for this model include a Hosmer-Lemeshow chi-square statistic of 7.943 (p-value = 0.439), Brier score = 0.127, and Nagelkerke pseudo R² = 0.569.

Discussion

We examined presentation, management, and outcomes of sepsis and septic shock at two tertiary referral hospitals in Rwanda. Mortality from sepsis without shock was 31.5%, and mortality from septic shock was 82.9%. The most common source of infection was intra-abdominal. Intravenous fluid resuscitation and antimicrobials were administered in most cases. Baseline characteristics at time of meeting sepsis criteria predicting mortality included respiratory rate, GCS score, and HIV+ status. When adjusting for these illness severity factors, management predictors of mortality were administration of vasopressors and mechanical ventilation.

Intravenous fluid resuscitation, antimicrobial administration, and source control are key aspects of sepsis management. In our sample, 93% received intravenous fluid resuscitation within 8 hours and the median volume received within this time frame was 1.0 L. The evidence for optimal standards of intravenous fluid resuscitation in sepsis in sub-Saharan Africa has been mixed [9–12]. In the only randomized controlled trial assessing intravenous fluid resuscitation in adults with sepsis in sub-Saharan Africa, Andrews et al. [10] found higher mortality in patients who received a greater volume of intravenous fluid resuscitation (3.5 L compared to 2.0 L within the first 6 hours after presentation). The Fluid Expansion As Supportive Therapy (FEAST) trial in African children revealed similar findings, as patients who received intravenous fluid bolus experienced higher mortality [18]. The most appropriate volume of fluid resuscitation in patients with sepsis remains unclear.

The vast majority of our sample (94.0%) received antimicrobial therapy. The most frequent antibiotic administered was ceftriaxone, and while we only had culture data from 8 patients in this study, all isolates tested for ceftriaxone were resistant to it. Our findings are similar to a 2018 study of peritonitis in Rwanda, in which 95.8% of patients with infectious peritonitis (95% meeting criteria for sepsis or severe sepsis) received antibiotics and the most frequently prescribed antibiotics were third generation cephalosporins (90% of cases) and metronidazole (85% of cases) [19]. Cephalosporin resistance was also high in that study as only one of seven isolates was sensitive to ceftriaxone. High rates of cephalosporin resistance in Rwanda has been demonstrated in several other studies as well [20–23]. Significantly more patients in our study received antimicrobials besides ceftriaxone and metronidazole later in their hospital course, presumably on clinical deterioration. Regarding source control, we collected data on intra-abdominal infections, and 43 out of 67 patients (64.2%) with an intra-abdominal source of infection underwent exploratory laparotomy.

Our findings of GCS, respiratory rate, and HIV+ status as significant baseline characteristics at time of meeting sepsis criteria predicting mortality have some overlap with Moore et al.’s [24] Universal Vital Assessment (UVA) score and Riviello et al.’s [4] Rwanda Mortality Probability Model (R-MPM). The UVA score, developed from data from 6 countries in sub-Saharan Africa, serves to predict mortality of patients in resource-limited settings admitted to the hospital for any reason. In addition to our above predictors that align with the UVA score, the score also includes temperature, heart rate, and systolic blood pressure. The R-MPM, based on data from patients admitted to the ICU, employs the mortality predictors of age, suspected or confirmed infection within 24 hours of ICU admission, hypotension or shock as a reason for ICU admission, Glasgow Coma Scale score (aligns with our findings) at ICU admission, and heart rate at ICU admission. In another study on peritonitis in Rwanda in which 83% of patients had sepsis or severe sepsis, Ndayizeye et al. [25] found that predictors of mortality included unplanned reoperation, vasopressor use (aligns with our findings), abnormal white blood cell count, ICU admission, and American Society of Anesthesiologist score of ≥ 3.

We found mechanical ventilation and administration of vasopressors to be the only management predictors of mortality. While it is possible that these management modalities do result in worse outcomes given risks of harm with these interventions, it is more likely that the use of mechanical ventilation or vasopressors are indicators of severity of illness that were not captured by our three-variable model. Further study is needed.

Our study had several limitations. First, we defined sepsis as suspected or confirmed infection plus two or more of the three qSOFA criteria (respiratory rate ≥ 22, systolic blood pressure ≤ 100, and Glasgow Coma Scale score < 15) [1]. The Sepsis-3 task force defined sepsis as “life-threatening organ dysfunction caused by a dysregulated host response to infection.” They suggested that the definition could be operationalized as an increase in SOFA score of 2 or more points, due to infection. They further suggested that the qSOFA criteria could be used to identify patients with a high likelihood of a poor outcome, particularly in out of hospital, emergency department, and ward settings [1]. Both operationalizations of the conceptual definition correlated with mortality. While the increase in SOFA score of two points is more commonly used to operationalize the definition for research, it was not possible in our cohort given that PaO2 and bilirubin, two key components of SOFA, are almost never available in this setting. We therefore used the qSOFA criteria to operationalize the definition of sepsis. It is not clear in what direction this might bias our results, but our population may be different than other cohorts that are defined using the SOFA score. We were also unable to include lactate in our definition of septic shock; again, this may mean that our cohort is somewhat different than others that define septic shock using both blood pressure and lactate. It is not clear in what direction this would bias our results.

Second, both our methods, as well as limitations in clinical documentation, led to a selection bias toward the inclusion of ICU patients over ward patients. While we screened all patients at CHUK discharged in 2017 for our inclusion criteria, at CHUB we screened all patients discharged in 2017 from the ICU, but only screened a portion of patients discharged from the internal medicine and surgery wards due to our resource limitations. This results in missing cases of sepsis at CHUB and a selection bias at CHUB toward patients discharged from the ICU. At both hospitals, vital signs are recorded with greater consistency in the ICU than in the wards, which may have led to additional selection bias toward ICU patients, since vital signs are included in the inclusion criteria. Finally, our inclusion criterion of two out of three qSOFA criteria, as opposed to one out of three criteria, may have resulted in bias toward more critically ill patients since every one point increase in qSOFA score is associated with higher mortality [26, 27]. We chose to use two out of three qSOFA criteria to achieve higher specificity [28, 29]. All of these factors lead to a selection bias that favors ICU patients being included in our study. This means that outcomes may be worse than might be expected from a comprehensive sampling of all hospital patients with sepsis.

Third, the limitations of clinical documentation for use in research led to an undercounting of sepsis cases, as well as missing data for the participants included. The CHUK central admission and discharge logbook lists final diagnosis or diagnoses as free text entries without standardization, and often with only primary diagnosis listed. It is possible that patients could have an infection or sepsis as one of their diagnoses, but that this would not be listed in the logbook. This may have led to fewer cases of sepsis at CHUK being captured. Indeed, previous studies have found higher proportions of hospitalized patients with a sepsis diagnosis [7, 30, 31], suggesting our method of screening may have missed patients with sepsis. Additionally, as this was a retrospective study, we were limited to data that was available in the medical records. Missing data rates ranged from 0% to 85.6%. This may have led to lower detection rates of sepsis since the three variables in the qSOFA score used for defining sepsis here include variables that are inconsistently recorded. This study cannot speak to the incidence or prevalence of sepsis at these facilities, given that our screening methods and inclusion criteria are likely to have missed cases of sepsis.

Missing values also mean that data on the included patients is incomplete in some cases. Laboratory values and imaging results are recorded inconsistently in physician documentation or are available on slips of paper which may be lost. Additionally, cardiac enzymes, liver and coagulation profiles, and arterial blood gasses are rarely ordered and thus our rates of myocardial infarction, liver failure, coagulopathy, and ARDS are almost certainly lower than actual rates. Patients may have had underlying comorbidities that were not documented in the medical record.

Fourth, we were limited in our ability to define septic shock using established criteria since serum lactate level is rarely available in our setting, and vasopressors are not consistently available to all patients given resource constraints. We were able to define septic shock only based on whether a vasopressor was administered, or whether a patient had hypotension. Fifth, we were unable to determine the timing of antimicrobial administration as the precise time is infrequently recorded.

Finally, while we attempted to control for severity of illness in the management mortality predictors model by including baseline characteristics that predicted mortality, we found nonetheless that receiving mechanical ventilation or vasopressor therapy remained as mortality predictors in the final model. The severity of illness scores are imperfect. The predictive value of mechanical ventilation and vasopressor therapy likely reflect the fact that more critically ill patients require these interventions, not that these interventions themselves are a cause of mortality.

The limitations of our study themselves suggest potential areas of quality improvement in sepsis care: monitoring and documentation of vital signs, as well as recognition of sepsis. Respiratory rate was missing in 38% of charts, and temperature in 33%. One-third of the subjects we identified as having sepsis did not have a sepsis diagnosis named in the medical record.

Future studies would greatly benefit from prospective data collection to document other aspects of care that may not have been recorded, to determine precise antimicrobial administration time and antimicrobial resistance patterns, and to assess potential delays in access to care and delays in operative procedures for source control, as these have been shown to be high and likely predict poor outcomes [32].

Conclusion

Sepsis and septic shock remain understudied in sub-Saharan Africa. Sepsis and septic shock mortality were high in our sample of patients at two tertiary referral hospitals. The vast majority of patients received therapy with antimicrobials and intravenous fluid resuscitation. Baseline characteristics that predicted mortality in our sample were GCS, respiratory rate, and HIV+ status. Future work should focus on optimal targets for intravenous fluid resuscitation, antimicrobial resistance and timing of administration, and appropriate use of vasopressors and mechanical ventilation.

Supporting information

Acknowledgments

We would like to thank Leroy R. Thacker PhD of Virginia Commonwealth University School of Medicine’s Department of Biostatistics for his statistical oversight and input. We also offer our thanks to Emmanuel Munyaneza MSc and Elias Rwamugema MSc for their assistance with accessing the paper medical record archives at CHUK and CHUB.