Numeric Result Cleaning

New Function : clean_numeric

The component nested functions and their impact on certain numeric fields are explained below:

Nested Function	Purpose
_additional_check_for_known_errors	CRP on admission, Ferritin and Troponin I: Extracts value from known error in "[6] - 2020-05-19" format. Extracts value if there is a leading "<" or ">". For example, '<4' becomes '4') Temperature on admission: Extracts value from string with additional information, for example, it returns '37.4' from '37.4 (36 hours post admission, 1st available)'.
_raise_if_date_in_numerical_column	During development one patient was found that has a date in the lymphocyte field (1/3/20) and another in Troponin T. Although only affecting a small number of values, this is raised for manual checking of the data.
_merge_o2_sat_into_po2	Only one value in 'o2_saturation' column, but this is now merged into PaO2/SpO2.
_split_and_clean_pao2	Where a centre has submitted a blood gas PaO2, this is converted from kPa to a percentage, then stored in 'pao2_gas'. Where a centre has submitted an SpO2, this is saved in 'pao2_saturation'
_impute_pao2_spo2	Imputes values of SpO2 from PaO2 values using the equation shown below.
_clean_creatinine	Creatinine data appears to have been submitted in multiple units (mg/dL and µmol/L). This function attempts to standardise the units to µmol/L.
_truncate_numerical_field	Truncates fields to account for the maximum and minimum values handled by the machines at some institutions. Further details are provided below.
_infer_ddimer_units	If a centre was not in the development data, the units for the D-Dimer are inferred by this function.
_clean_d_dimer	All values are converted to ng/mL FEU.
_merge_Ferritin_2	Ferritin_2 column is merged with 'Ferritin'.
_check_sheffield_trops	The units specified by Sheffield to NHSx for their Troponin I values do not fit with those in the data. This function warns the developer to check these values are consistent with those of other centres and correct/exclude this data accordingly.

---

Significantly Impacted Fields

PaO₂ / SpO₂

Note

PaO₂ < 7 is likely venous - these are removed by default.

• PaO₂ versus SpO₂:

  • When requesting data, NHSx asked for PaO₂ (Partial Pressure of Oxygen) with the vital signs rather than the more commonly collected SpO₂ (Oxygen Saturation). As a result, some centres submitted PaO₂’s from arterial blood gases (ABGs) and others SpO₂’s (from a pulse oximeter).

  • The typical units of PaO₂ (kPa) and SpO₂ (%) are a factor of 10 different, allowing differentiation between centres which submitted a PaO₂ and an SpO₂.

  • 3 hospitals completed this field in kPa exclusively.

• Other corrections:

  • Royal United Hospitals Bath: 1. Put some FiO₂’s in the wrong column (e.g. 0.21, 21, 26, 38, 50); and 2. Entered some blood gas values (PaO₂).

  • Ealing Hospital and Ashford and St Peters appear to have entered FiO₂ values.

  • Oxford University Hospitals and Liverpool Heart and Chest Hospital have entered SpO₂ as a fraction rather than a percentage, i.e. 0.XX where XX is the SpO₂ as a percentage.

• Solutions:

  • Integer values between and including 21-50 are taken as FiO₂ if FiO₂ is blank. If FiO₂ appears blank a warning is raised.

  • Any common oxygen fractions <=0.5, e.g. 0.21 are assumed to be FiO₂ if FiO₂ is blank. If FiO₂ is not blank a warning is raised.

  • Any values for that 0.5<=PaO₂<=1 holds are assumed to be SpO₂ as a fraction of 1 and are multiplied by 100.

  • The PaO₂ column is then split into pao2_gas and spo2_saturation

  • SpO₂ is imputed from PaO₂ values to merge the columns using the following equation[1]:

    \[SpO_2 = \left(\frac{28.6025^3}{{PaO_{2}}^{3}}+0.99 \right)^{-1} \]

---

Creatinine

• A handful of values are less than 20 and unlikely to be in SI units (µmol/L):

  • Some are less than 0.5 and appear to be in mmol/L rather than µmol/L.

  • Others between 0.5 and 20.0 are decimals (and therefore unlikely to be µmol/L), but too large for mmol/L. These could be errors (e.g. decimal placement) or in mg/dL; however, they did not appear consistent with mg/dL. These values are clipped in the clip_numeric function.

---

D-Dimer

• D-Dimer may be in DDU (D-Dimer Units) or FEU (Fibrinogen Equivalent Units). The standard is now FEU and use of this was confirmed with the labs.

• Some centres had values orders of magnitude different from the others as they used any of: ng/mL, μg/mL, mg/L, g/L. For all of the centres in the development data, this was checked by telephone.

• At some centres, it was apparent their machine had a maximum possible value above which results were truncated (e.g. if a centre had multiple results at a maximum of exactly 10,000). The minimum maximum laboratory value identified was 10,000, and consequently results were truncated to this value as if all machines had this as their maximum.

---

Truncated fields

Note

D-dimer and Ferritin have their minimum values clipped to 0 rather than truncated, as negative values are clearly errors (see Clip Numeric Fields).

Feature	Minimum	Maximum
crp_on_admission	4
d-dimer_on_admission		10,000
ferritin		15,000
troponin_i	10
troponin_t	5	25,000

---

[1]

Gadrey, S., Lau, C., Clay, R., Rhodes, G., Lake, D., Moore, C. & Voss, J. & Moorman, J. (2019). Imputation of partial pressures of arterial oxygen using oximetry and its impact on sepsis diagnosis. Physiological Measurement. 40. 10.1088/1361-6579/ab5154.