Phoenix Sepsis EHR Implementation Notes

The Phoenix criteria for pediatric sepsis and septic shock were developed using an international data set of pediatric emergency department and hospital encounters. Birth hospitalizations and patients with a gestational age less than 37 weeks were excluded, please see below (Sanchez-Pinto et al. 2024; Schlapbach et al. 2024). The Phoenix criteria were developed based on predicting encounter mortality among those with suspected infection using organ dysfunction-based data from the first 24 hours of the encounter. Please see the original manuscripts for details.

The purpose of this document is to provide details about the expected inputs for the scoring in order to assist EHR implementation. The phoenix R package and Python module, along with the example SQL code assumes that the data have been prepared as expected and returns scores accordingly.

General Data Formatting Notes

NULL values are treated as non-score generating criteria. Do not populate null/unknown/not applicable values with estimates or outdated information.
Values outside the ranges mentioned below are treated as NULL
Before sending data to the package, please verify the data. Some values may provide warnings, others may generate errors. Verify that your code can handle warnings and errors.
Ideally, both input values and time of the measurement are available so that last observation carried forward logic can be implemented with limits and conditions for age of inputs.

Inclusion/Exclusion criteria

Inclusion Criteria

Child < 18 years old
Suspected infection

Exclusion Criteria

Patients still in their original birth hospitalization
Gestational age less than 37 weeks at encounter start
Age greater than or equal to 18 years (216 months) at encounter start

Suspected Infection

For a patient to have sepsis or septic shock based on the Phoenix criteria, they must have a suspected infection. A patient is considered to have a suspected infection if they have:

a test for infection ordered, and
at least two doses of systemic antimicrobial medication(s).

Note that the test, e.g., a blood culture, need only be ordered. Positive/negative test status is not needed.

Systemic antimicrobial medications are any antibiotic, antiviral, or antifungal medication which had any of the following routes:

ASSUMED SYSTEMIC
ENTERAL
HEMODIALYSIS
INTRA-ARTERIAL
INTRACARDIAC
INTRADUODENAL
INTRAGASTRIC
INTRAILEAL
INTRAMUSCULAR
INTRAVASCULAR
INTRAVENOUS
INTRAVENOUS BOLUS
INTRAVENOUS DRIP
NASOGASTRIC
ORAL
OROPHARYNGEAL
PARENTERAL
RECTAL
SUBCUTANEOUS

Other non-systemic routes such as intraocular, nasal, otic, etc. should be excluded.

Example SQL code used for mapping medication routes when developing Phoenix can be found in the project GitHub archive specifically the files: med_route_mapping.sql and clean_up_routes.sql.

General notes for all subscores

Reasonable values for observations and labs used during development of the Phoenix Sepsis Score are documented in file standard_names_and_units.csv within the project GitHub archive.

“Look Back” for each input below indicates the length of time a last value should be carried forward.

Respiratory

Respiratory dysfunction scoring is based on PaO₂/FiO₂ (PFR), SpO₂/FiO₂ (SFR), and respiratory support. While the scoring is based on SFR and PFR, it is preferable to use separate SpO₂, PaO₂, and FiO₂ values so data checks can be performed before the score is determined.

Inputs

PaO₂: Partial pressure of oxygen in arterial blood

Units: mmHg
Expected Values: [0, Inf)
Caveats:
Look Back: 6 hours

SpO₂: Peripheral oxygen saturation (pulse ox)

Units: percentage
Expected Values: [0, 100]
- Expect integer values 0, 1, 2, …, 100
Caveats: values greater than 97 should not be used for SFR
Look back: 6 hours

FiO₂: Percent inspired oxygen

Units: not applicable
Expected Values: [0.21, 1.00]
- expected floating point
- 0.21 is room air
- 1.00 is pure oxygen
Caveats:
Look back: 6 hours

PFR: PaO₂/FiO₂

Units: mmHg
Expected Values [0, Inf)
- Expect floating point

Caveats:

PFR is only valid if the time of FiO₂ value is the same as, or before, the time of PaO₂ value. For example, if FiO₂ was reported at 11:31 and PaO₂ was reported at 11:45 then the PFR would be valid to use. If the FiO₂ is then updated at 11:59, the PaO₂ value from 11:45 would not be valid to use in the calculation of the PFR

* New value
^c value carried forward
time	FiO₂	PaO₂	PFR	Note:
11:31	0.30*
11:45	0.30^c	118*	393.33
11:47	0.30^c	118^c	393.33
11:59	0.40*	118^c		No PFR, FiO₂ is newer than PaO₂
12:10	0.40^c	152*	380
12:15	0.40^c	152^c	380
17:45	0.40^c	152^c	380
18:00		152^c		No PFR, FiO₂ carry forward window expired
18:30				No PFR, FiO₂ and PaO₂ carry forward windows have expired

Look back: implied by the look back on PaO₂ and FiO₂ and the condition that FiO₂ value is the same age or older than the PaO₂ value.

SFR: SpO₂ / FiO₂

Units: none
Expected Values: [0, 97]/0.21
- Floating point
- The upper limit includes the limitation on SpO₂ ≤ 97

Caveats:

SFR is only valid if the time of FiO₂ value is the same as, or before, the time of SpO₂ value. For example, if FiO₂ was reported at 11:31 and SpO₂ was reported at 11:45 then the SFR would be valid to use. If the FiO₂ is then updated at 11:59, the SpO₂ value from 11:45 would not be valid to use in the calculation of the SFR

* New value
^c value carried forward
time	FiO₂	SpO₂	SFR	Note:
10:15	0.25*	92*	368
10:20	0.25^c	98*		No SFR, SpO₂ > 97
11:31	0.30*	98^c		No SFR, SpO₂ > 97 and FiO₂ newer than SpO₂
11:45	0.30^c	87*	290
11:47	0.30^c	87^c	290
11:59	0.40*	87^c		No SFR, FiO₂ is newer than SpO₂
12:10	0.40^c	85*	212.5
12:15	0.40^c	85^c	212.5
17:45	0.40^c	85^c	212.5
18:00		85^c		No SFR, FiO₂ carry forward window expired
18:30				No SFR, FiO₂ and SpO₂ carry forward windows have expired

Look back: implied by the look back on SpO₂ and FiO₂ and the condition that FiO₂ value is the same age or older than the SpO₂ value.

ORS: Other Respiratory Support

Units: not applicable
Expected Values:
- 0: No other respiratory support
- 1: some form of respiratory support; may or may not include supplemental oxygen
Caveats:
- If FiO₂ > 0.21 then that indicates oxygen support and thus ORS
Look Back: 6 hours
- If it has been more than six hours since last known data point indicating the patient has respiratory support then consider the patient to no longer have respiratory support.

IMV: Invasive Mechanical ventilation

Units: not applicable
Expected Values:
- 0: patient does not currently have invasive mechanical ventilation
- 1: patient currently has invasive mechanical ventilation
Caveats:
- IMV due to surgery should not be used to assess organ dysfunction
Look back: 6 hours
- If it has been more than 6 hours since the last known data point indicating the patient has IMV, then consider the patient to no longer have IMV

Scoring

Scoring is based on the lowest available PaO₂/FiO₂ (PFR) and SpO₂/FiO₂ (SFR) with consideration for IMV and ORS.

# R code: booleans are implicitly coerced to integers
cat(tail(as.character(body(phoenix::phoenix_respiratory)), 1), sep = "\n")
#> imv * (((pfr < 100) | (sfr < 148)) + ((pfr < 200) | (sfr < 220))) + ors * ((pfr < 400) | (sfr < 292))

Cardiovascular

Cardiovascular dysfunction is based on three conceptual sets:

Systemic Vasoactive medications
Lactate
Age adjusted mean arterial pressure

Inputs

Systemic Vasoactive Medications

There are six systemic vasoactive medications to consider:

dobutamine
dopamine
epinephrine
norepinephrine
milrinone
vasopressin

Vasoactive medications (dobutamine, dopamine, etc.) should only count if they are systemic. Ideally one would identify flowsheet or medication rows where the dose is documented every hour for each infusion. In our dataset, epinephrine had some non-systemic medication administrations that needed to be excluded. Here are some non-systemic examples to exclude:

epinephrine 0.3 mg/0.3ml inj soln prefilled syringe
epinephrine hcl (nasal) 0.1 % nasal solution
epinephrine hcl 2.25 % inh neb soln

The routes for administration of the vasoactive medications are at Example SQL code used for mapping routes when developing Phoenix can be found in the project GitHub archive specifically the files: med_route_mapping.sql and clean_up_routes.sql.

The overall metadata for vasoactives:

Units: none
Expected Values: integer values 0, 1, 2, 3, 4, 5, 6
- (maximum vasoactive points is reached with 2)
Caveats: see above
Look back: 12 hours

Lactate

Units: mmol/L
Expected Values: [0, 50]
Caveats:
Look back: 6 hours

Age

Units: Months
Expected Values: [0, 216)
Caveats:
Look back:

Mean Arterial Pressure (MAP)

There can be many sources of blood pressure. We used the following hierarchy for which value to use in the scoring.

Arterial MAP (invasive measurement)
Estimated MAP from arterial systolic (SBP) and arterial diastolic (DBP) pressures
Cuff MAP
Estimated MAP from cuff SBP and cuff DBP

MAP can be estimated as (2/3) * DBP + (1/3) * SBP

Expected metadata:

MAP:
- Units: mmHg
- Expected Values: [1, 300]
- Caveats:
  - If MAP is provided along with SBP and DBP then verify that DBP ≤ MAP ≤ SBP
- Look back: 6 hours
SBP:
- Units: mmHg
- Expected Values: [1, 300]
- Caveats:
  - Verify that DBP ≤ SBP, we used a slightly more conservative check of (SBP - DBP) ≥ 1
- Look back: 6 hours
DBP:
- Units: mmHg
- Expected Values: [1, 200]
- Caveats:
  - Verify that DBP ≤ SBP, we used a slightly more conservative check of (SBP - DBP) ≥ 1
- Look back: 6 hours

Scoring

Up to six points can be earned from cardiovascular dysfunction. Up to two points for vasoactives, up to two points for lactate, and up to two points for MAP.

# vas: number of vasoactives meds
# lct: lactate in mmol/L
# age: age in months
# map: mean arterial pressure in mmHg
cat(tail(as.character(body(phoenix::phoenix_cardiovascular)), 4), sep = "\n")
#> vas_score <- (vas > 1) + (vas > 0)
#> lct_score <- (lct >= 11) + (lct >= 5)
#> map_score <- ((age >= 0 & age < 1) * ((map < 17) + (map < 31)) + (age >= 1 & age < 12) * ((map < 25) + (map < 39)) + (age >= 12 & age < 24) * ((map < 31) + (map < 44)) + (age >= 24 & age < 60) * ((map < 32) + (map < 45)) + (age >= 60 & age < 144) * ((map < 36) + (map < 49)) + (age >= 144 & age <= 216) * ((map < 38) + (map < 52)))
#> vas_score + lct_score + map_score

Coagulation

Inputs

Platelets

Units: 1000/μL
Expected Values: [0, Inf)
Caveats:
Look back: 24 hours

INR

Units: none
Expected Values: [0, Inf)
Caveats:
Look back: 24 hours

D-dimer

Units: mg/L FEU
Expected Values: [0, 500]
Caveats:
Look back: 24 hours

Fibrinogen

Units: mg/dL
Expected Values: [0, Inf)
Caveats:
Look back: 24 hours

Scoring

While there are four inputs and a patient could get a point for each one, the final score is capped at two points.

cat(tail(as.character(body(phoenix::phoenix_coagulation)), 2), sep = "\n")
#> rtn <- as.integer(plt < 100) + as.integer(inr > 1.3) + as.integer(ddm > 2) + as.integer(fib < 100)
#> pmin(rtn, 2)

Neurologic

Inputs

Glasgow Coma Scale (GCS)

Phoenix is based on the total GCS score.

GCS (total):
- Units: none
- Expected Values: integer values 3, 4, 5, …, 15
- Caveats:
  - The total GCS is the sum of three component scores: eye, motor, and verbal.
  - We assumed that if we had an updated score for any one of the three components but not all, then the non-updated components were assumed to have not changed. In that case, we updated the time of the GCS total and all its components to the most recent update time for any of the components.
- Look back: 12 hours
GCS (eye):
- Units: none
- Expected Values: integer values 1, 2, 3, 4
- Caveats:
- Look back: 12 hours
GCS (motor):
- Units: none
- Expected Values: integer values 1, 2, 3, 4, 5, 6
- Caveats:
- Look back: 12 hours
GCS (verbal):
- Units: none
- Expected Values: integer values 1, 2, 3, 4, 5
- Caveats:
- Look back: 12 hours

Pupils

Units: none
Expected Values:
- 1: bilaterally fixed pupils
- 0: one or zero fixed pupils
Caveats:
Look back: 6 hours

Scoring

At most 2 points.
1 point for GCS ≤ 10
2 points for bilaterally fixed pupils

A patient with a GCS ≤ 10 and bilaterally fixed pupils will have a score of 2.

# fpl: 1 if bilaterally fixed pupils; 0 otherwise
cat(tail(as.character(body(phoenix::phoenix_neurologic)), 1), sep = "\n")
#> pmin(fpl * 2 + as.integer(gcs <= 10), 2)

Additional inputs to Phoenix-8 score

Endocrine

Inputs

Glucose

Units: mg/dL
Expected Values: [5, 2000]
- floating point
Caveats:
Look back: 12 hours

Scoring

Scoring for Phoenix-8 only. 1 point if glucose < 50 or > 150

cat(tail(as.character(body(phoenix::phoenix_endocrine)), 1), sep = "\n")
#> as.integer((glc < 50) | (glc > 150))

Immunologic

Inputs

ANC

Units: 1000 cells / mm³
Expected Values: [0, Inf)
Caveats:
- Be careful with the units. The R package, Python module, and example SQL code expect units in 1000 cells / mm³. Tables and other references might use cells / mm³
Look back: 24 hours

ALC

Units: 1000 cells / mm³
Expected Values: [0, Inf)
Caveats:
- Be careful with the units. The R package, Python module, and example SQL code expect units in 1000 cells / mm³. Tables and other references might use cells / mm³
Look back: 24 hours

Scoring

At most one point

cat(tail(as.character(body(phoenix::phoenix_immunologic)), 1), sep = "\n")
#> as.integer((anc < 0.5) | (alc < 1))

Renal

Inputs

Age (see notes in cardiovascular)

Creatinine

Units: mg/dL
Expected Values: [0, 50]
Caveats:
Look back: 24 hours

Scoring

Age-adjusted creatinine

cat(tail(as.character(body(phoenix::phoenix_renal)), 1), sep = "\n")
#> (age >= 0 & age < 1) * (crt >= 0.8) + (age >= 1 & age < 12) * (crt >= 0.3) + (age >= 12 & age < 24) * (crt >= 0.4) + (age >= 24 & age < 60) * (crt >= 0.6) + (age >= 60 & age < 144) * (crt >= 0.7) + (age >= 144 & age <= 216) * (crt >= 1)

Hepatic

Inputs

Total Bilirubin

Units: mg/dL
Expected Values: [0, 100]
- floating point
Caveats:
Look back: 24 hours

ALT

Units: IU/L
Expected Values: [0, Inf)
Caveats:
Look back: 24 hours

Scoring

0 or 1 points

cat(tail(as.character(body(phoenix::phoenix_hepatic)), 1), sep = "\n")
#> as.integer((bil >= 4) | (alt > 102))

References

Sanchez-Pinto, L. Nelson, Tellen D. Bennett, Peter E. DeWitt, Seth Russell, Margaret N. Rebull, Blake Martin, Samuel Akech, et al. 2024. “Development and Validation of the Phoenix Criteria for Pediatric Sepsis and Septic Shock.” JAMA, January. https://doi.org/10.1001/jama.2024.0196.

Schlapbach, Luregn J., R. Scott Watson, Lauren R. Sorce, Andrew C. Argent, Kusum Menon, Mark W. Hall, Samuel Akech, et al. 2024. “International Consensus Criteria for Pediatric Sepsis and Septic Shock.” JAMA, January. https://doi.org/10.1001/jama.2024.0179.

THIS IS A WORK IN PROGRESS

General Data Formatting Notes

Inclusion/Exclusion criteria

Suspected Infection

General notes for all subscores

Respiratory

Inputs

PaO2: Partial pressure of oxygen in arterial blood

SpO2: Peripheral oxygen saturation (pulse ox)

FiO2: Percent inspired oxygen

PFR: PaO2/FiO2

SFR: SpO2 / FiO2

ORS: Other Respiratory Support

IMV: Invasive Mechanical ventilation

Scoring

Cardiovascular

Inputs

Systemic Vasoactive Medications

Lactate

Age

Mean Arterial Pressure (MAP)

Scoring

Coagulation

Inputs

Platelets

INR

D-dimer

Fibrinogen

Scoring

Neurologic

Inputs

Glasgow Coma Scale (GCS)

Pupils

Scoring

Additional inputs to Phoenix-8 score

Endocrine

Inputs

Glucose

Scoring

Immunologic

Inputs

ANC

ALC

Scoring

Renal

Inputs

Age (see notes in cardiovascular)

Creatinine

Scoring

Hepatic

Inputs

Total Bilirubin

ALT

Scoring

References

PaO₂: Partial pressure of oxygen in arterial blood

SpO₂: Peripheral oxygen saturation (pulse ox)

FiO₂: Percent inspired oxygen

PFR: PaO₂/FiO₂

SFR: SpO₂ / FiO₂