Emergency Care of Sepsis in Sub-Saharan Africa: Incidence, Mortality and Non-Physician Clinician Management of Sepsis in rural Uganda from 2010 to 2019

Data were obtained for patients seen from 2010-2019 in a rural Ugandan emergency unit staffed by non-physician clinicians. Sepsis was dened as suspected infection with a qSOFA score ≥ 2. Descriptive analysis was performed and a multi-variable logistic regression mortality model was created. Analysis included Wilcoxon rank-sum test, t-test, one-way ANOVA, and Fisher’s exact test.

emergency medicine is rapidly developing -Non-physician clinicians provide emergency care in Uganda but the quality and outcomes of their care are unknown What are the new ndings?
Sepsis represents a large burden of emergency care in rural Uganda in the largest published study of sepsis mortality outcomes published to date in sub-Saharan Africa covering the decade from 2010 -2019 Sepsis and especially malarial sepsis incidence decreased during the study period and non-physician clinicians provided increasingly high quality of care yet sepsis mortality rates increased An increased relative risk of death was associated with all sepsis interventions including uids, antibiotics or both in retrospective multiple variable logistic regression suggesting confounding by indication What do the new ndings imply?
Emergency unit trends mirror national and regional trends in sepsis over the last decade Even large scale retrospective analysis is unable to de ne optimal sepsis management and prospective randomized controlled trials are necessary Background Sepsis is de ned as life-threatening organ dysfunction caused by a dysregulated host response to infection, while septic shock is a subset of sepsis in which there are underlying circulatory, cellular, and metabolic abnormalities that are associated with an even greater risk of mortality. [1] Sepsis is one of the most signi cant worldwide causes of morbidity and mortality, with 48.9 million annual cases and 11 million annual deaths (19.7% of all global deaths) according to the most recent analysis of the Global Burden of Disease Study from 2017 data. [2] Sepsis disproportionately affects persons in low-and-middle income countries (LMICs). [2,3] The World Health Organization has identi ed sepsis as an international priority, adopting a resolution in 2017 to improve the prevention, diagnosis, and clinical management of sepsis. [4] While in-hospital mortality rates can be greater than 25-30% across resource settings, mortality rates of up to 38% are reported in LMICs, particularly in patients who have con rmed bacteremia. [1,3,5] In Uganda, a low-income country in sub-Saharan Africa (SSA), the primary causes of death are all infectious in nature -namely malaria, HIV/AIDs, pneumonia, tuberculosis and diarrhea -and all can be associated with sepsis. Inpatient mortality rates have been reported between 34-43%. [6][7][8][9] Identifying patients with sepsis and septic shock presents a signi cant challenge. Clinical criteria to identify patients with these syndromes have evolved over time. The most recent Sepsis-3 consensus recommendations suggest using bedside clinical criteria, termed the quick Sequential Organ Failure Assessment (qSOFA) score. [1] Adult patients with suspected infection and two or more of these ndings are determined to meet sepsis criteria. Using the qSOFA score in a low-resource setting is particularly useful because it does not rely on advanced diagnostic testing. 10 It is a concise, objective scale which provides opportunity for a rapid objective clinical evaluation of the patient by a variety of providers including nurses, mid-level providers, and physicians.
Management of sepsis globally has changed signi cantly over the past two decades. In the US and other high-resource settings, "early goal-directed therapy" sparked widespread interest and change in practice patterns for sepsis management, while subsequent trials have called into question the need for higherresourced aspects of the protocol. [10][11][12][13][14][15] To add to confusion, a landmark study in children in SSA showed that uid bolus, a cornerstone of management in high resource settings, might actually increase mortality.
[16] Some subsequent studies showed similar increased mortality with uids. [17,18] However, most experts agree that management should focus early antibiotics and source control plus some combination of uid resuscitation and early initiation of vasopressor therapy, tailored to the patient presentation.
Emergency medicine is rapidly developing in SSA generally and Uganda speci cally. [19][20][21] With this development, initial sepsis care is shifting to the emergency unit and a better understanding of the burden and outcomes of sepsis from an emergency medicine perspective is required. Sepsis guidelines speci c to emergency care in SSA are rare, with the African Federation of Emergency Medicine Handbook serving as the most de nitive reference with its recommendations for uids and early antibiotics following the Surviving Sepsis Campaign. [22] The following analysis utilizes rural Ugandan emergency unit data collected over the decade from 2010 to 2019 and describes sepsis incidence, care and mortality. This evidence base for emergency unit sepsis care will contribute to understanding of larger trends in sepsis incidence over time in Uganda, the ability of non-physician clinician training programmes to impact emergency care quality, and the impact of sepsis management on mortality.

Description of Study Site
All data comes from the emergency unit at Karoli Lwanga Hospital, a rural district hospital located in the town of Nyakibale in the Rukungiri District of southwest Uganda. The hospital has a six-bed emergency unit that opened in 2008 and treats 300 to 700 patients per month arriving between 8:00 am and midnight every day of the year. Since 2010, the emergency unit has been staffed independently by nonphysician clinicians called emergency care practitioners (ECPs) who received training from physicians working with Global Emergency Care, a Uganda-and US-based non-governmental organization. The ECPs are nurses who have completed a two-year advanced training course in emergency care described in detail elsewhere. [20] While training from 2008 to 2010 was initially supervised by visiting US emergency medicine physicians, ECPs have practiced independent care unsupervised by physicians since 2010 and with subsequent training of new cadres in the pilot phase of the project. There is not any real-time physician supervision of clinical care, although the ECPs consult local physicians for patients who require surgery, do not respond to initial treatments, or in whom there is considerable diagnostic uncertainty.
Emergency unit patients were admitted to medical and surgical wards staffed by Ugandan physicians with standard levels of training and no connection to Global Emergency Care.
During the study period, the hospital lacked critical care units, ventilators, capabilities for invasive monitoring, and vasopressor medications, other than epinephrine vials for intended management of anaphylaxis. Diagnostic testing was limited to hemoglobin, blood grouping, urinalysis, blood smear for malaria, glucose, cerebrospinal uid analysis, and HIV rapid testing along with radiology (X-ray and ultrasound) of variable availability. Over time, some additional testing including basic metabolic panel became available and ECPs adopted bedside ultrasonography. Resource utilization by clinicians in this emergency unit is described in detail elsewhere. [23] Global Emergency Care maintained a prospectively collected quality assurance database of all Karoli Lwanga Hospital emergency unit patient visits, including demographics, vital signs, clinician impression of patient clinical status, medications administered and procedures performed, as well as follow-up vital status (mortality) for all admitted and discharged patients. On Day 3 following initial evaluation in the emergency unit, patients admitted to the hospital were visited in person, while patients discharged from the emergency unit or ward were contacted via phone (if the patient has a phone). A rigorous follow-up protocol which included seven consecutive days of calling patients on the phone before considering them lost to follow-up was used for all patients and this protocol is described in detail elsewhere. [20] When the database was developed, three-day follow-up was chosen to minimize loss to follow-up for admitted and discharged patients in a setting where most patients do not have consistent ability to receive phone calls. Additionally, follow-up after three days was thought to be less re ective of outcomes related to acute care provided in the emergency unit. Ethics approval for the quality assurance database was obtained through the Institutional Review Board at Mbarara University of Science and Technology. Data was input during patient stays in the emergency unit by Global Emergency Care-trained research assistants present in the emergency unit. They

Data Analysis
Retrospective analysis was performed on prospectively collected data abstracted from the Karoli Lwanga Hospital emergency unit quality assurance database, including all consecutive patients presenting to the emergency unit from January 2010 until December 2019. Suspected infections were de ned by a complaint of fever, an objective fever, or a diagnosis consistent with infection (full list of diagnosis is available as Additional le 1 ). Sepsis was de ned as a patient with a suspected infection and a qSOFA score of two points or greater with one point each for: tachypnea (respiratory rate ≥ 22); altered mentation (de ned as a Glasgow Coma Scale <15 or an AVPU score other than Alert); hypotension (systolic blood pressure ≤ 100). Hypotension in sepsis was de ned as sepsis with a mean arterial pressure less than or equal to 65. After consultation with clinicians, patients without recorded mental status were assumed to have a mental status of "Alert" for analysis, based on practice patterns of typically omitting recording mental status if it was normal. No imputation was otherwise performed for missing data. Malaria was de ned as either "smear-positive" (a positive thick/thin smear for Plasmodium falciparum), or "clinical" (patient received diagnosis or treatment of malaria prior to arrival and/or clinical suspicion was high enough despite a negative thick/thin smear). Demographics, vital signs, administration of anti-infectives (including antimalarial, antiviral, antifungal and/or antibiotic medicines) and intravenous uids (normal saline and/or Ringer's lactate solution), malaria testing (), HIV status (previously known or tested in the emergency unit), and three-day mortality outcomes were analyzed for all patients. Non-parametric age data were compared using the Wilcoxon rank-sum test; continuous variables were compared using the t-test and one-way ANOVA; proportions were compared using Fisher's exact test. A multi-variable logistic regression model to test the signi cance of associations between independent variables and mortality in septic patients was developed. Area Under Receiver Operating Characteristics Curve (AUROC), Hosmer-Lemeshow Goodness of Fit, and Brier score were all calculated for this model.

Patient and public involvement
The non-physician clinician training programme was originally developed in response to several years of clinical emergency medicine experience and ongoing health care sta ng shortages in Uganda. The positive response of patients, staff and administrators at Karoli Lwanga Hospital to the training programme and their interest in improving patient care led to ongoing research and programme evaluation. Patients and the public were not involved in the design of the study however outcome measures are explicitly patient-oriented. Results will be disseminated through open access publication to allow local clinicians, administrators, policy-makers and researchers to bene t.

Results
Overall, there were 48,653 patient visits from 2010-2019 ( Figure 1). Of these, 17,490 patients were included in the analysis who were aged 18 and older and had a suspected infection. In total, 3,323 patients were de ned as having sepsis based on qSOFA scores of 2 or greater. Loss to follow-up analysis was strati ed by disposition as the emergency unit overall experienced fundamentally different rates of loss to follow-up for admitted (n=702 of 11,927, 5.9%) and discharged patients (n=2,748 of 5,403, 50.9%).
Admitted patients were visited in person by research assistants, and follow-up for discharged patients relied on patients owning and answering mobile phones. When comparing septic and non-septic patients by disposition, admitted patients in both groups had similar loss to follow up (5.8% vs. 6.0%, p=0.75), while discharged patients with sepsis were less likely to be lost to follow up than non-septic patients (n=128 of 340, 37.7% vs. n=2,620 of 5,063, 51.8%, p<0.001). The overall mortality rate for combined Page 7/24 septic and non-septic discharged patients with follow-up was low with only 4 deaths (n=4 of 2,657, 0.15%) over the 10 years of the study.
The annual number of visits for all patients, patients with suspected infections and with malarial and non-malarial sepsis are displayed in Figure 2. The prevalence of sepsis for patients aged 18 years or older during the study period was 10.4% for all emergency unit visits. The annual proportion of all visits with suspected infections decreased from 61.7% in 2010 to 51.0% in 2019, with the proportion of those with suspected infections and meeting sepsis criteria decreasing from 16.4% in 2010 to 4.7% in 2019. Nonmalarial sepsis decreased from 12.1% in 2010 to 4.6% in 2019. Malarial sepsis (combined "smearpositive" and "clinical") had an even more pronounced decrease from its peak of 6.5% in 2013 to 0.3% in 2019. The proportions in Figure 2 were calculated using all emergency unit patient visits as a denominator (n=31,856) to describe overall emergency unit trends. All subsequent analyses were restricted to patients with suspected infection (n=17,490).
Characteristics of septic and non-septic patients with suspected infection were compared (Table 1). There were signi cant differences between the groups, with septic patients being younger, more likely to be female, more likely to have abnormalities in all vital signs (blood pressure, heart rate, respiratory rate, oxygen saturation, mental status), more likely to be coinfected both with HIV and malaria, more likely to have a clinician impression of "Serious" or "Critical" illness, and more likely both to be admitted to the hospital and to die in the emergency unit. Levels of missing data were generally low: age (n=110, 0.63%), gender (n=12, 0.07%), blood pressure (n=681, 3.89%), heart rate (n=515, 2.94%), respiratory rate (n=1,460, 8.35%), oxygen saturation (n=1,338, 7.65%), and clinician impression of illness severity (n=118, 0.67%).
Mental status was an exception with a high level of missing data (n=10,720, 62.7%).
The mortality and interventions in patients strati ed by qSOFA score are displayed in Table 2 below. Interventions were incompletely captured for the 3,942 patients in the original Excel database from 2010 -2012, therefore analysis of interventions in sepsis over time was restricted to the 13,548 patients recorded in the Access database from 2012 -2019. Increasing qSOFA scores were associated with higher rates of death and higher rates of receiving both uids and anti-infectives. The dichotomous splitting of patients into "No Sepsis" and "Sepsis" using the cutoff of qSOFA≥2 shows the same clinically and statistically signi cant associations.  Figure 4 displays trends in prevalence and associated mortality for sub-groups of septic patients by year from 2010-2019. Across programme years, septic patients are more likely to be elderly and to have qSOFA scores of 3 (both higher mortality sub-groups), less likely to have malaria (a lower mortality sub-group) and equally likely to have hypotension (a much higher mortality sub-group).
To understand the relative impact of these changes in both septic patients and sepsis care, a multivariable logistic regression model was created for septic patients. Patients with complete treatment data from 2012 -2019 both with malaria (n=741) and without malaria (n=1621) were separated to identify independent predictors of mortality. Looking at malarial sepsis only, there were 54 deaths (7.3% mortality rate) in the 741 cases from 2012-2019. Further analysis of this data over time was limited by the precipitous drop in malaria incidence and deaths culminating in the years 2017-2019 having 31, 26 and 11 cases and two, three and one deaths respectively. Analysis of the impact of treatments was limited by the baseline group of "No treatment received in ED" having only one death in 156 cases, thus making comparisons unusably imprecise. Consequently, malaria was excluded from subsequent logistic regression analysis in this manuscript but is available as Additional le 2. Looking at the multiple variable logistic regression model for non-malarial sepsis mortality (Table 3), the odds ratio (OR) of death was signi cantly increased seen in patients with advancing age, hypothermia, hypotension, tachycardia, with both hypoxia and tachypnea, and for patients with concerning clinical condition upon arrival ("Sick" and "Toxic"). There was a signi cantly reduced OR of death for patients who were female or febrile. The p-value for the Hosmer-Lemeshow goodness of t test was 0.10, the Brier score was 0.079, and the AUROC was 0.80 (95%CI 0.77 -0.83). The predicted mortality coe cients generated by this model were plotted against observed mortality across programme years and both predicted and observed mortality could be seen to increase across the study period ( Figure 5).
The categorical variable for sepsis treatment in the emergency unit displayed visually in Figure 3a above (no treatment in the emergency unit, uids alone, antibiotics alone, or both uids and antibiotics) was added to the logistic regression model above to assess the independent association between treatments and mortality in all septic patients without malaria. All treatments for sepsis were associated with increased relative risk (RR) of death as compared to "no treatment in the emergency unit" after controlling for the variables in the logistic regression model above. The RR were as follows: " uids alone" RR=1. and Risk Factors (GBD) survey. 6 The National Malaria Control Programme in Uganda which was active during the study period, employed a combination of control measures including long-lasting insecticidal nets, indoor residual spraying, and intermittent preventive treatment for malaria during pregnancy and its success is the likely cause of the observed reduction in emergency unit burden. 29 The sharp downward trend in non-malarial sepsis and sepsis overall seen here has also been observed during the study period in national and regional level analyses of GBD data. [2,25] While GBD data is not speci c to the emergency unit, the few emergency medicine-focused studies from SSA show a similar downward trend in sepsis incidence.
[26] The ultimate causes of this trend are likely tied to overall economic and health systems development in Uganda and throughout SSA in general, but detailed analysis of the impact of those forces on emergency unit malaria incidence and mortality is outside the scope of this manuscript.
The data presented in Table 2 showed that qSOFA performed well in classifying patients with higher and lower mortality rates in line with other research suggesting the value of qSOFA in SSA. [27] There was clear identi cation of increased mortality not just between septic/non-septic patients (qSOFA<2 and ≥2), but also an increase of crude mortality as qSOFA scores increased from zero to three. Overall this analysis both contributes to support for qSOFA in low-income, SSA emergency unit settings and represents the rst study to the authors' knowledge that demonstrates the potential utility of qSOFA in a setting where clinical care is provided by non-physician clinicians. Further prospective studies would be needed to fully validate the utility of qSOFA by non-physician clinicians.
Non-physician clinician sepsis care quality also improved signi cantly during the study period (as depicted in Table 3a). The non-physician clinician training employed in the study setting promotes ongoing practice improvement and has focused on increasing the rates of early resuscitation of septic patients in accordance with regional guidelines for emergency care. [22] The rates of uid and antibiotic administration provided by non-physician clinicians compared favorably to the standard of care provided by admitting medical o cers in other Ugandan hospitals during the same time period. [9] Some authors have suggested that improved training of nurses, paramedical assistants, and other non-physician clinicians could signi cantly improve sepsis identi cation and management and can be done at low cost.
[6, 28] Taking sepsis care as a reasonable proxy for emergency care in general, the ndings of this manuscript have clear policy implications: emergency medicine training programmes for non-physician clinicians produce a workforce capable of delivering quality care, as de ned by meeting published care guidelines. Such programmes may address both emergency care sta ng shortages in SSA and help meet the pressing global development goals of improving quality of care. [29] Despite the increasing quality of care, observed mortality for septic patients increased across the study period while mortality for non-septic patients did not (as seen in Table 3b). However, the prevalence of higher and lower mortality subgroups of septic patients also changed substantially over time, which was hypothesized to have changed the predicted mortality over time (as seen in Figure 4). When both predicted and observed mortality were plotted over time in Figure 5, it was evident that both did in fact increase over time. Given the global trends towards reduced sepsis incidence and mortality cited above, this was somewhat surprising. One possible explanation is "self-triage" or the decisions made by patients about where to seek emergency care for themselves or their family. Self-triage can be an important factor in many settings and the local providers and patients have given feedback to the development programme for many years the local community has viewed the emergency unit positively since it was established in 2008. One interpretation of the gradually increasing predicted mortality across programme years is that the local community's increasing con dence in the emergency unit over the years has led increasingly sick patients to preferentially seek care there. This interpretation is supported by the data displayed in Figure 4 showing a marked increase in the proportions of patients in high-mortality subgroups (e.g. elderly, qSOFA=3) over time. While these variables were included in the overall mortality model, further analysis of this phenomena is outside the scope of this manuscript.
One variable not commonly included in mortality models -but which independently predicted increased mortality in this model -was the initial "clinical impression" of patients upon their arrival to the emergency unit. The training programme has long stressed that the clinical skill of rapidly identifying both critically-ill and generally well-appearing patients is a cornerstone of emergency care. Every nonclinician trained by the programme was taught to categorize patients as "not sick", "sick" or "toxic" early in their evaluation and prioritize interventions based on this assessment. The strongly signi cant independent association of this variable with mortality ("Sick" OR = 2.2, 95%CI 1.2 -3.9; "Toxic" OR = 6.7, 95%CI 3.2 -13.8) even when controlling for comorbidities, age and vital sign abnormalities suggests that this type of clinical assessment has utility in addition to other objective data in the emergency unit. The ability of emergency care-trained non-physician clinicians to utilize this skill central to emergency medicine physician training to reliably identify patients at higher risk for short-term mortality contributes to the overall argument for their clinical capacity to evaluate and provide unsupervised emergency care for undifferentiated septic patients.
Despite the clearly increasing adherence to quality care guidelines across time (as de ned by providing uids and anti-infectives in the emergency unit and depicted in Figure 3a) there was no trend towards reducing observed versus predicted mortality and in fact both observed and predicted mortality increased during the study period (Figure 3b, Figure 5). Furthermore, the logistic regression model including these interventions showed increased mortality with all sepsis interventions and requires additional investigation. This apparent harm is noteworthy as optimal sepsis resuscitation in SSA remains controversial with a recent review of sepsis guidelines in SSA citing multiple studies showing harm from uids. [16,18,30,31] Retrospective analysis such as that used in this study is challenged by one of the rst principles of emergency medicine: the most critically-ill patients with the highest mortality rates receive the most interventions. Table 2 provides supporting data for the existence of this care pattern by demonstrating signi cant associations between increasing interventions, increasing mortality and increasing qSOFA scores. This practice pattern ultimately creates substantial confounding by indication in associating interventions with mortality without clear causation.
The logistic regression model for mortality was developed in an attempt to control for confounding by indication by including not only objective data (e.g., vital signs) but also subjective data (e.g., clinical impression). It was hoped that the inclusion of subjective evaluation would control for the recognition of critical illness by the providers and their associated decisions to escalate interventions. However, this model ultimately showed a clinically but not statistically signi cant association between mortality and either uids (RR=1.22, 95%CI 0.57 -1.87) or antibiotics (RR=1.25, 95%CI 0.60 -1.91) given singly, and a clinically and statistically signi cant association between mortality and uids and antibiotics given together (RR=1.85, 95%CI 1.02 -2.69). No literature published to date suggests that antibiotics increase mortality in sepsis in any setting. Therefore, their association with increased mortality functions as a falsi cation test arguing that the model was unable to adequately control for confounding and arguing against drawing any conclusions regarding causative associations between treatments for sepsis and mortality in this model. Different analytic approaches may provide additional answers but, given that this manuscript includes a decade of data and more septic patients than exist in all other studies of sepsis outcomes in SSA combined, retrospective analysis may be inadequate to de ne optimal sepsis care in Uganda or SSA due to this overwhelming confounding by indication. As health systems and emergency care rapidly develop in Uganda and SSA more generally, they do so without the necessary evidence base to guide optimal sepsis care. Controlled, prospective, randomized clinical trials are needed to isolate the mortality impact of individual sepsis treatments and/or care bundles for emergency unit patients in these settings.

Limitations
There are several limitations for this study. The rst limitations are those of the registry database. Loss to follow-up for discharged patients was high, despite rigorous methods that included calling patients every day for seven consecutive days following discharge. However, the large number of patients that did not have phones or answer their phones limited response rates. Despite mortality being exceedingly low at 0.15% over 10 years of discharged patients, this low response rate likely resulted in underestimating discharged mortality. Multiple imputation of mortality data to control for this was considered but deemed methodologically unsound. Secondly, the registry data are limited to a single center. The regional speci city of causes and management of sepsis may limit the generalizability of the ndings above.
Mortality reported in this study was lower overall than in similar settings. This may be due to larger trends in mortality in the region but also likely re ects limitations of the emergency unit registry data which only recorded vital status at three days and therefore may underestimate longer-term measures such as total in-hospital mortality. Unique patient visits were recorded and reported instead of unique patients which may impact overall incidence of sepsis. As discussed above, logistic regression models were likely limited in their ability to control for confounding by indication. Limitations in data prevented inclusion weightbased dosing of uids and appropriateness of anti-infectives as variables in analysis. Future randomized prospective studies may bene t from looking speci cally at sepsis outcomes related to types of antiinfectives and volume of uids administered. Missing data regarding mental status (omitted in over 60% of charts) may have systematically under-recognized sepsis based on qSOFA.

Conclusions
Data for 3,323 septic patients from 2010-2019 at a single rural Ugandan emergency unit staffed by nonphysician clinicians were included in retrospective analysis and represent the largest study of sepsis outcomes in SSA published to date. The annual proportions of patients with sepsis decreased over time, and the non-physician clinicians providing emergency care signi cantly improved their adherence to sepsis guidelines. These ndings provide insight into how sepsis in Ugandan emergency medicine re ects sepsis trends at national and regional levels and demonstrate that non-physician clinicians are both capable of delivering high-quality emergency care in SSA and represent a possible solution to ongoing healthcare sta ng shortages. As the nature of patients seeking emergency sepsis care changed over time the expected mortality rates increased but, despite improvements in care quality, the observed mortality rates also increased. Ultimately, logistic regression models were unable to control for confounding and isolate the mortality impact of uids and/or antibiotics in sepsis. With retrospective analysis unable to address ongoing concerns about the safety of uids in sepsis in SSA and emergency medicine developing rapidly in both Uganda and SSA, randomized controlled trials are urgently needed to de ne and guide optimal sepsis management.
Abbreviations CI -con dence interval ECP -emergency care practitioner LMICs -low-and middle-income countries OR -odds ratio qSOFA -quick Sequential Organ Failure Assessment RR -relative risk SSA -Sub-Saharan Africa Declarations Ethics approval and consent to participate Ethics approval for the quality assurance emergency unit database was obtained through the Institutional Review Board at Mbarara University of Science and Technology and University of Massachusetts.

Consent for publication (individual)
Not applicable.

Availability of data and materials
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare no competing interests.  Trends in prevalence and associated mortality for sub-groups of septic patients by year from 2010-2019.