Evaluation of Organ Dysfunction Scores for Allocation of Scarce Resources in Critically Ill Children and Adults during a Healthcare Crisis

Abstract

Objective:

When healthcare systems are overwhelmed, accurate assessments of patients’ predicted mortality risks are needed to ensure effective allocation of scarce resources. Organ dysfunction scores can serve this essential role, but their evaluation in this context has been limited so far. In this study, we sought to assess the performance of three organ dysfunction scores in both critically ill adults and children at clinically-relevant mortality thresholds and timeframes for resource allocation and compare it to two published prioritization schemas.

Design, Setting and Patients:

Retrospective observational cohort study of critically ill patients in three large academic medical centers in the US.

Interventions:

None.

Measurements and Main Results:

We calculated the daily Sequential Organ Failure Assessment score (SOFA) in adults and the Pediatric Logistic Organ Dysfunction 2 (PELOD-2) score and the Pediatric SOFA (pSOFA) score in children. There were 49,290 adults (11.6% mortality), and 19,983 children (2.5% mortality) included in the analysis. Both the SOFA and pSOFA scores had adequate discrimination across relevant timeframes and adequate distribution across relevant mortality thresholds. Additionally, we found that the only published state prioritization schema that includes pediatric and adult patients had poor alignment of mortality risks, giving adults a systematic advantage over children.

Conclusions:

In the largest analysis of organ dysfunction scores in a general population of critically ill adults and children to date, we found that both the SOFA and pSOFA scores had adequate performance across relevant mortality thresholds and timeframes for resource allocation. Published prioritization schemas that include both pediatric and adult patients may put children at a disadvantage. Furthermore, the distribution of patient and mortality risk in the published schemas may not adequately stratify patients for some high-stakes allocation decisions. This information may be useful to bioethicists, healthcare leaders, and policy makers who are developing resource allocation policies for critically ill patients.

INTRODUCTION

The coronavirus disease 2019 (COVID-19) pandemic has highlighted the limited surge capabilities of critical care resources in the United States (1). Despite previous models predicting a shortage of critical care beds and mechanical ventilators during viral pandemics, there were few efforts to develop contingency plans and policies for the allocation of critical resources prior to COVID-19 (2, 3). While the current pandemic will eventually subside, this will certainly not be the last crisis to strain the healthcare system and challenge the delivery of critical care services (4).

In a national healthcare crisis, the ethical responsibility of clinicians to care for individual patients is often superseded by public health policies prioritizing “the greatest good for the greatest number of patients,” yet how that goal is achieved is often unclear, and policies and guidelines vary between institutions (1, 5, 6). Many guidelines call for a multi-principled approach wherein prioritization for patients most likely to survive hospitalization is considered along with other criteria such as the patient’s age and the presence of life-limiting comorbidities (1, 2, 7, 8). Examples of published prioritization schemas in the medical literature include the ones currently used in the states of Maryland and Pennsylvania (1, 2).

Almost all allocation guidelines and prioritization schemas incorporate measures of short-term survival and most use the Sequential Organ Failure Assessment (SOFA) score to determine the risk for in-hospital mortality in critically ill adult patients (1, 2, 4, 7, 9, 10). While there are other severity of illness scores available, the SOFA score has several advantages: it is easy to calculate with a limited set of laboratory variables, it has been validated in a wide variety of critical care conditions, and it has seen increased use since the publication of the most recent consensus definitions of sepsis (9, 11–13). However, the SOFA score is designed for adult patients. At the time of our writing, only one published schema in the medical literature has incorporated a pediatric organ dysfunction score alongside the adult SOFA score (2). Furthermore, the empirical evaluation of these organ dysfunction scores in the context of care prioritization in the United States has been limited, so the performance of these tools as a component of a multi-principled prioritization schema is unknown. While many prioritization schemas consider other principles, such as the likelihood of long-term survival in the setting of comorbidities and the prioritization of patients who have experienced fewer years of life, the most commonly recommended preliminary assessment to allocate scarce resources relies on the likelihood of survival to discharge as determined by organ dysfunction scores, favored because of their relative objectivity and avoidance of unfairly disadvantaging vulnerable populations (1, 2, 4, 7–10).

In this study, we focused on this key component of prioritization schemas: the likelihood of surviving the hospitalization. We empirically assessed the performance and characteristics of organ dysfunctions scores to discriminate in-hospital mortality in critically ill children and adults using large, multi-center, electronic health record databases while considering key aspects relevant to healthcare crises situations and assessing the score ranges proposed in published prioritization schemas. Our goal is to help bioethicists, healthcare leaders, and policy makers develop allocation guidelines for critical care resources during different types of healthcare crises while maintaining transparency related to the likelihood of survival to discharge as a key component of prioritization schemas.

METHODS

Study design and data sources

We conducted a retrospective observational cohort study of critically ill children and adults admitted to three large academic medical centers in the United States. The adult population included all patients 18 years or older admitted to the Beth Israel Deaconess Medical Center in Boston, Massachusetts between 2001 and 2012, who are part of a publicly available intensive care unit (ICU) database (14). The pediatric population included all children younger than 18 years of age admitted to the pediatric ICUs at the University of Chicago Comer Children’s Hospital and the Ann & Robert H. Lurie Children’s Hospital of Chicago between 2010 and 2016. Data were extracted from the three respective databases using structured queries and underwent quality checks for conformity, completeness, and plausibility (15). For practical reasons, only the information pertaining to the first ICU encounter and up to 28 days, discharge or death (whichever came first) in each hospitalization was included, as well as whether the patient survived the hospitalization or not. Each admission was treated independently.

Organ dysfunction scores and primary outcome

While the use of the SOFA score in adults is well established in many resource allocation and prioritization schemas, there has been limited use of pediatric scores alongside adult scores for this purpose, with the exception of the Pediatric Logistic Organ Dysfunction Score version 2 (PELOD-2) score in one schema published in the medical literature (2). However, in a previous single-center study we have shown that a pediatric version of the SOFA score (pSOFA) has comparable performance to the PELOD-2 score in critically ill children with the added advantage of requiring fewer variables (16). Furthermore, pSOFA variables are similar to SOFA variables, making implementation in crisis situations easier. Therefore, we calculated the SOFA score in adult patients, and the PELOD-2 and the pSOFA scores in children for comparison purposes. We calculated the organ dysfunctions scores for every 24-hour period starting at ICU admission using previously published criteria (16–18). Additional details on the calculation of the scores can be found in the Supplemental Digital Content. The primary outcome was in-hospital mortality.

Discrimination and distribution of the organ dysfunction scores

We studied the discrimination of in-hospital mortality of the three organ dysfunction scores on day 1 of admission, as well as the maximum score reached within 3 days and 28 days of ICU admission. Day 1 was chosen to evaluate the performance of the scores in situations of triaging and early resource allocation. The 3-day timeframe was chosen based on previously proposed timeframes for re-evaluation of critically ill patients in crises situations (4). Finally, the 28-day timeframe was chosen to account for the vast majority of the organ dysfunction burden expected later in critically ill patients. Collectively, our study evaluates the performance of the score in important situations during a healthcare crises with limited resources, such as: (i) immediate resource allocation as well as re-allocation that may occur after the first few days (e.g. renal replacement therapy, extra-corporeal life support, etc.); (ii) as a prognostic tool to consider for end-of-life decision-making in the highest risk patients ; and (iii) as a throughput aid to accelerate discharge and relocation to lower acuity areas for low risk patients.

Additionally, we assessed the performance of the organ dysfunction score in the subset of mechanically ventilated patients, given that ventilators are one of the potentially most scarce resources in healthcare crises, especially those due to respiratory viral pandemics like COVID-19 or influenza. Finally, we examined the discrimination of in-hospital mortality by the SOFA score when calculated using the peripheral oxygen saturation-to-fraction of inspired oxygen (SpO₂:FiO₂) ratio in addition to the PaO₂:FiO₂ ratio for the respiratory dysfunction subscore, given that in healthcare crises, particularly those with high respiratory disease burden, a non-invasive measure of oxygenation may be easier to obtain (4, 9, 19).

The calibration of the scores were studied by analyzing the distribution of patients at three clinically-relevant thresholds of in-hospital mortality: 5%, 25%, and 50%. The 5% threshold was chosen to characterize a low risk group of patients, potentially eligible for early transfer out of the ICU or triage to off-site care facilities (4, 5). The 25% and 50% mortality thresholds were chosen to mirror mortality rates from previously published adult SOFA-based prioritization schemas and to capture a substantial mortality difference between groups (e.g. >25% difference as proposed by Hick et al. (10)), as these groups likely represent patients in which the most critical re-allocation of resources will take place.

Evaluation of the organ dysfunction scores in existing prioritization schemas

We studied the distribution of the SOFA and PELOD-2 scores and mortalities associated based on the categories proposed in two published schemas. We termed these prioritization schemas the “Maryland” and “Pennsylvania” schemas based on their intended regional use (1, 2). The Maryland schema includes scores ranges for both adult and pediatric patients (SOFA and PELOD-2), whereas the Pennsylvania schema includes only score ranges for adult patients using the SOFA score.

While the use of organ dysfunction scores to assess the likelihood of surviving the hospitalization is a key component of these two prioritization schemas, it is not the only component. Both of these schemas use a multipronged approach in which other principles are considered, such as the likelihood of long-term survival in the setting of comorbidities and the prioritization of patients who have experienced fewer years of life (or life cycles) (1, 2). However, for the purposes of this analysis, we only focused on the likelihood to survive the hospitalization as determined by the organ dysfunction scores.

Statistical analyses

Data were analyzed using R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria). Categorical variables were compared using the chi-squared test and continuous variables using the non-parametric Kruskal-Wallis test. In-hospital mortality discrimination was assessed using the area under the receiver operating characteristic curve (AUC) and compared using the DeLong method. To study the relationship between organ dysfunction scores, age, and mortality, patients were classified into four pediatric and four adult age groups (1, 8).

The Institutional Review Boards at Ann & Robert H. Lurie Children’s Hospital of Chicago and The University of Chicago approved this study with a waiver of informed consent.

RESULTS

Out of 69,273 ICU admissions included in the analysis, 49,290 were adults (38,426 unique patients) with 11.6% in-hospital mortality, and 19,983 were children (13,913 unique patients) with 2.5% in-hospital mortality. Demographics, critical care support, and organ dysfunction score distributions are presented for pediatric and adult patients and compared across survivor and non-survivors of the hospitalization. In both adult and children, non-survivors required more organ support and had higher organ dysfunctions scores. In the pediatric cohort, non-survivors were more commonly younger, whereas in the adult cohort older patients had higher mortality. Finally, male gender was associated with mortality only in the adult cohort.

Discrimination and distribution of the organ dysfunction scores

SOFA score in adult patients

The SOFA score had adequate discrimination of in-hospital mortality with AUCs ranging from 0.70 on day 1 to 0.77 by day 28 in all patients, and 0.70 to 0.73 in mechanically ventilated patients (Supplemental Digital Content Table 1). The discrimination was consistently better when adding the SpO₂:FiO₂ ratio as an alternative for the respiratory score instead of using the PaO₂:FiO₂ ratio alone (p<0.001, Supplemental Digital Content Table 1).

The median day of the maximum SOFA score was on day 1 (i.e. the first 24 hours) for all mortality groups, except for the >50% mortality group in which it was day 2 (Supplemental Digital Content Table 2). The median time to death of non-survivors was 4.1 days from admission (interquartile range [IQR] 1.4–9.1 days) and 2.1 days after the peak SOFA score (IQR 0.9–5.8 days).

The ranges, frequency, and distribution of the maximum SOFA score by day 28 across the 5-25-50% mortality thresholds are presented in Figure 1 and Table 2. Additionally, the data presented in Figure 1 can be interacted with using the data visualization tool associated with this article (available at https://nsanchezpinto.shinyapps.io/allocation_app/). In the online data visualization tool, readers may select different timeframes and score ranges and perform sensitivity analyses based on age groups and mechanical ventilation use. Finally, the mortality across adult age groups in the different score ranges is presented in Figure 2.

Figure 1. Range, frequency, and distribution of the maximum Sequential Organ Failure Assessment (SOFA) score by day 28 across clinically-relevant mortality thresholds in critically ill adults.

The top figure represents the in-hospital mortality and its 95% confidence interval (error bars) across the individual scores. The horizontal dashed lines represent the mortality thresholds at 5%, 25%, 50%. The bottom figure represents the distribution of patients reaching each score during their ICU admission. The vertical dashed lines represent the different groups based on the mortality thresholds. The percentages represent the proportion of ICU patients in each group. The data presented in this figure can be interacted with using the data visualization tool associated with this article (available at https://nsanchezpinto.shinyapps.io/allocation_app/).

Table 2.

Distribution of patients and mortality associated with the organ dysfunction score ranges in the three schemas.

Schema	SOFA score				PELOD-2 score				pSOFA score
5-25-50% Mortality Thresholds	Score range	Distribution, No. %	Died, No. %		Score range	Distribution, No. %	Died, No. %		Score range	Distribution, No. %	Died, No. %
	0 to 3	22483 (45.6)	787 (3.5)		0 to 10	18553 (92.8)	124 (0.7)		0 to 8	17952 (89.8)	105 (0.6)
	4 to 8	20934 (42.5)	2584 (12.3)		11 to 15	1016 (5.1)	137 (13.5)		9 to 12	1390 (7.0)	139 (10.0)
	9 to 12	4451 (9)	1437 (32.2)		16 to 17	162 (0.8)	59 (36.4)		13 to 18	572 (2.9)	205 (35.8)
	>12	1422 (2.9)	908 (63.9)		>17	252 (1.3)	174 (69.0)		>18	69 (0.3)	45 (65.2)

	SOFA score				PELOD-2 score
Maryland1	Score range	Distribution, No. %	Died, No. %		Score range	Distribution, No. %	Died, No. %
	0 to 8	43417 (88.1)	3371 (7.8)		0 to 11	18868 (94.4)	141 (0.7)
	9 to 11	3837 (7.8)	1149 (29.9)		12 to 13	420 (2.1)	59 (14.0)
	12 to 14	1372 (2.8)	704 (51.3)		14 to 16	380 (1.9)	95 (25.0)
	>14	664 (1.3)	488 (73.5)		>16	315 (1.6)	199 (63.2)

	SOFA score
Pennsylvania2	Score range	Distribution, No. %	Died, No. %
	0 to 5	34000 (69)	1938 (5.7)
	6 to 8	9417 (19.1)	1433 (19.1)
	9 to 11	3837 (7.8)	1149 (29.9)
	>11	2036 (4.1)	1174 (58.5)

¹The Maryland SOFA score ranges are based on the schema proposed by Daugherty et al. (2).

²The Pennsylvania SOFA score ranges are based on the schema proposed by White et al. (1). The Maryland and Pennsylvania allocation protocols assign priority scores from lowest to highest priority based on four different organ dysfunction score ranges, represented by the rows in the table, e.g. 0 to 8, 9 to 11, etc. for the SOFA score in the Maryland protocol. In addition, both protocols also assign priority scores based on the presence of major or severe comorbidities. In both cases, a patient who has an organ dysfunction score in a lower range will have higher priority for resource allocation given the higher likelihood of survival to discharge.

Abbreviations: SOFA, Sequential Organ Failure Assessment; PELOD-2, Pediatric Logistic Organ Dysfunction version 2; pSOFA, Pediatric SOFA.

Figure 2. Mortality associated with different age groups and the pSOFA/SOFA score mortality groups in pediatric and adult patients.

Abbreviations: SOFA, Sequential Organ Failure Assessment; pSOFA, Pediatric SOFA.

pSOFA and PELOD-2 scores in children

The pSOFA score had good-to-excellent discrimination of in-hospital mortality with AUCs ranging from 0.87 on day 1 to 0.93 by day 28 in all patients, and 0.86 to 0.90 in mechanically ventilated patients (Supplemental Digital Content Table 3). The discrimination was better than the PELOD-2 score on day 1 and by day 3 in all patients (p≤0.01) and similar by day 28 and in mechanically ventilated patients (p≥0.14, Supplemental Digital Content Table 3).

The median day of the maximum pSOFA score was day 1 for all mortality groups, except for the >50% mortality group in which it was day 4. The median day of the maximum PELOD-2 score was day 1 for the <5% and >50% mortality groups, and day 2 for the other groups (Supplemental Digital Content Table 2). The median time to death of non-survivors was 4 days from admission (interquartile range [IQR] 1.3–17.7 days) and 2.3 days after the peak pSOFA or PELOD-2 score (IQR 1–11.2 days).

The ranges, frequency, and distribution of the PELOD-2 and pSOFA score across the 5-25-50% mortality thresholds are presented in Figures 2 and Table 2. Overall, the pSOFA score had slightly better distribution of patient across the mortality threshold groups than the PELOD-2 score. When comparing it to adults, children had much lower mortality overall and, as expected, were less likely to be in high risk categories. For example, ~40% of adult patients fell into the 5–25% mortality range, whereas only 5–7% of children fell into that same range (Table 2). Additionally, the data presented in Figure 3 can be interacted with using the data visualization tool (available at https://nsanchezpinto.shinyapps.io/allocation_app/). Finally, the mortality across pediatric age groups in the different score ranges is presented in Figure 2.

Figure 3. Range, frequency, and distribution of the maximum Pediatric Sequential Organ Failure Assessment (pSOFA) score (left) and Pediatric Logistic Organ Dysfunction version 2 (PELOD-2) score (right) by day 28 across clinically-relevant mortality thresholds in critically ill children.

The top figures represent the in-hospital mortality and its 95% confidence interval (error bars) across the individual scores. The horizontal dashed lines represent the mortality thresholds at 5%, 25%, 50%. The bottom figures represent the distribution of patients reaching each score during their ICU admission. The vertical dashed lines represent the different groups based on the mortality thresholds. The percentages represent the proportion of ICU patients in each group. The data presented in this figure can be interacted with using the data visualization tool associated with this article (available at https://nsanchezpinto.shinyapps.io/allocation_app/).

Evaluation of the organ dysfunction scores in existing prioritization schemas

The Maryland schema showed poor alignment in mortality across pediatric and adult patients. The in-hospital mortality associated with the pediatric PELOD-2 score categories in the Maryland schema was consistently lower than the adult SOFA score categories, giving adults a systematic advantage over children (Table 2). For example, adult patients in the second tier had an average mortality of 29.9%, whereas children in the same tier had an average mortality of 14%. Furthermore, the mortality differences amongst the tiered groups in the adult patients in the schema were in no case >25%, which was proposed by Hicks et al. as a substantial difference for resource re-allocation (10).

The Pennsylvania schema, targeted for adults only, classified the majority of patients (69%) into the lower tier (SOFA score ≤5), and these had an average in-hospital mortality of 5.7%. The highest tier (SOFA score >11) accounted for 4.1% of patients, and these had an average in-hospital mortality of 58.5%, which was approximately 30% higher than the average mortality of the third tier group (Table 2). Of note, the average mortality difference between the second and third tier groups was only 11%, which means that on average prioritization between these two groups may be relying on a difference in risk of in-hospital mortality which may not be substantial (e.g. >25%) depending on what the prioritization is for (e.g. renal replacement therapy in patients with fluid overload or ventilator allocation in patients with respiratory failure).

DISCUSSION

In this study we present the largest analysis to date of organ dysfunction scores in a general population of critically ill patients across all age groups, focusing on aspects of the scoring systems that are relevant for resource allocation in in different types of healthcare crisis situations (1, 2, 9). We evaluated the performance characteristics, ranges, and distribution of the SOFA, PELOD-2, and pSOFA scores across three clinically-relevant mortality thresholds and timeframes as well as in two previously published prioritization schemas (1, 2).

The performance characteristics, frequency, and mortality distribution of the SOFA score in our analysis of almost 50,000 adult ICU patients is very similar to the findings by Raith and colleagues in over 180,000 adult ICU patients with confirmed or suspected bacterial infection (11). Similarly, the PELOD-2 and pSOFA score were consistent with previously published studies in critically ill children, both with and without infections (20). This demonstrates a high degree of consistency of the scores across populations, which is a desirable characteristic for objective measures of short-term mortality risk (8).

One of the key findings of our study is that the only prioritization schema published in the medical literature that includes organ dysfunction scores for both children and adults -the Maryland schema- had poor alignment of the mortalities across the two populations in our study (2). As currently designed, the ranges of the PELOD-2 scores have the potential to disadvantage children with moderate-to-high mortality risk by giving them the same priority as adults with twice their mortality risk. Importantly, this prioritization schema is currently in use in two states for ventilator allocation during the COVID-19 pandemic. Four other states that include a pediatric score in their schemas use the original PELOD score, which is now outdated, has been shown to have poor calibration, and has not been compared or aligned with the adult SOFA score (6, 17, 21).

There are several justifications for including critically ill children in prioritization schemas alongside adult patients and for ensuring that there is adequate alignment of mortality risks. Children draw from the same pool of resources as adults, including ventilators and other life support equipment, medications, hospital beds, blood products, and protective equipment. Additionally, while children may take priority based on the principle of maximizing years of life, prioritization based on that principle is still controversial, with differing opinions both amongst experts and community members (7). Finally, while children have been relatively spared from the burden of COVID-19, it is conceivable that future healthcare crises will involve patients across all age groups, including other respiratory viral pandemics (for example, the 1918 influenza pandemic disproportionately affected children and adults less than 40 years old (22)), a combination of COVID-19 in adults and another severe respiratory viral pandemic in children, or as a consequence of natural or man-made disasters. Therefore, children should indeed be included in prioritization schemas, acknowledging that they have increased likelihood of survival at baseline and therefore have a built-in advantage, as fewer of them will reach the high-risk categories of well-aligned prioritization schemas, as demonstrated by our results.

Our study is important for several reasons. The current COVID-19 pandemic has highlighted the limited preparedness and surge capabilities of critical care services in the United States (1). Most hospitals are using prioritization guidelines developed after the 2009 H1N1 influenza pandemic, but there have been important advances since then (4, 8). The availability of large electronic health record databases now allows us to empirically test the performance of organ dysfunction scores in general ICU populations of both children and adults, including testing new pediatric scores like the pSOFA, and alternatives to existing scores, such as the use of the SpO₂:FiO₂ ratio to calculate the SOFA score. Furthermore, a detailed description of the distribution of the scores across mortality ranges can help bioethicists, healthcare leaders, and policy makers develop prioritization schemas across age groups with full transparency related to the risk of mortality. This is particularly important for institutions without a prioritization schema or for those that do not include criteria for pediatric patients. In a recent COVID-19 international survey of pediatric ICUs, only 44 of 91 (48%) institutions had a bioethics team or policy to determine allocation of beds or other resources (23). Importantly, of those with a policy, 84% used the likelihood to survive the acute episode as top criteria for prioritization, further emphasizing the need for transparency in the performance of organ dysfunctions scores for this purpose.

Our findings are subject to several limitations. First, the data for this study was collected retrospectively with the adult data covering the years 2001–2012. However, the distribution of mortality rates across SOFA scores in this study were consistent with more recent studies, thereby illustrating the consistency of SOFA scores in assessing mortality across populations and time periods (11). Second, our study population included general ICU patients mostly during non-crisis situations, so the distributions of the scores reflect a normal baseline population and not what may be expected during a pandemic. Further studies are required to translate these findings to different scenarios during healthcare crises. Third, while the organ dysfunction scores have good performance at the population level, their performance in individual patients is bound to vary. However, we show that these scores can be used as a prognostic enrichment strategy in groups of patients for prioritization purposes. Lastly, while the SOFA and pSOFA appear to have adequate performance in general populations of critically ill patients and when considering key aspects relevant to healthcare crises situations, they are not without limitations. For example, some of the score components may have some built-in subjectivity, including the assessment of GCS or the timing for initiation of vasoactive medications. Future studies may help identify more objective scores that overcome these limitations.

CONCLUSION

In this study we present the largest analysis to date of organ dysfunction scores in general populations of both adult and pediatric critically ill patients using the SOFA, pSOFA, and PELOD-2 scores. We show that these scores, particularly the SOFA score in adults and the pSOFA score in children, have good discrimination of short-term survival and adequate distribution of patients across meaningful mortality thresholds and timeframes relevant to resource allocation. Importantly, we found that published prioritization schemas that include both pediatric and adult patients may put children at a disadvantage. Additionally, the distribution of patient and mortality risk in the published schemas may not adequately stratify patients for some high-stakes allocation decisions. This information may be useful to bioethicists, healthcare leaders, and policy makers who are developing policies for resource allocation in healthcare crises.

Table 1.

Demographics, critical care support required, and organ dysfunction scores in critically ill children and adults.

	Non-Survivors	Survivors	p-value
Pediatric Patients
Total No.	494	19,489
Age Groups, No. (%)
<1 year	132 (26.7)	3,809 (19.5)	0.001
1–4 years	131 (26.5)	6,153 (31.6)
5–11 years	115 (23.3)	4,800 (24.6)
12–18 years	116 (23.5)	4,727 (24.3)
Male (%)	281 (56.9)	10,698 (54.9)	0.41
Required mechanical ventilation, No. (%)	430 (87)	6701 (34.4)	<0.001
Required vasoactive infusions, No. (%)	350 (70.9)	1318 (6.8)	<0.001
pSOFA Scores, median (IQR)
Maximum by day 3	11 (8, 14)	3 (1, 5)	<0.001
Maximum by day 28	13 (9, 15)	3 (1, 5)	<0.001
PELOD-2 Scores, median (IQR)
Maximum by day 3	13 (7, 18)	3 (2, 5)	<0.001
Maximum by day 28	15 (10, 19)	3 (2, 6)	<0.001
Adult Patients
Total No.	5,712	43,578
Age Groups, No. (%)
18–40 years	284 (5)	4,999 (11.5)	<0.001
41–60 years	1,208 (21.1)	13,272 (30.5)
61–74 years	1,514 (26.5)	12,613 (28.9)
>74 years	2,706 (47.4)	12,694 (29.1)
Male (%)	3,095 (54.2)	24,554 (56.3)	0.002
Required mechanical ventilation, No. (%)	2,934 (51.4)	9,995 (22.9)	<0.001
Required vasoactive infusions, No. (%)	2,847(49.8)	6,831 (15.7)	<0.001
SOFA Scores, median (IQR)
Maximum by day 3	6 (4, 10)	3 (2, 5)	<0.001
Maximum by day 28	7 (5, 11)	3 (2, 5)	<0.001

Abbreviations: IQR, interquartile range; PELOD-2, Pediatric Logistic Organ Dysfunction version 2; pSOFA, Pediatric Sequential Organ Failure Assessment.