fdiamant / ckd_mimiciv

Towards a CKD (Chronic Kidney Disease) dataset.

---

Creating a dataset for Chronic Kidney Disease

---

Aim

This mini project stems from my master's thesis conducted at Stockholm University^1,2. The primary focus is on creating a CKD (Chronic Kidney Disease) dataset from the MIMIC-IV (Medical Information Mart for Intensive Care) [1] database while identifying potential issues within the dataset and proposing possible solutions to mitigate them. The secondary aim is to develop an understanding of the eGFR (estimated Glomerular Filtration Rate) calculation process and contribute to enhancing the accuracy and reliability of the results. The insights gained from this project can be valuable for future researchers seeking to extract information on kidney health or other health related data projects from the dataset.

eGFR and the CKD-EPI(2021) equation

The acronym eGFR refers to estimated Glomerular Filtration Rate, which serves as a cost-effective and accurate method for evaluating kidney function. Depending on the eGFR value and other factors beyond the scope of this project, healthcare professionals may recommend lifestyle modifications and/or prescribe specialized medications tailored to each individual patient.The National Kidney Foundation offers an informative fact sheet³ and a CKD heat map⁴ documents that provide a comprehensive introduction to eGFR and Chronic Kidney Disease (CKD).

For the most up-to-date calculation of eGFR, researchers have proposed the CKD-EPI(2021) equation [2]. The National Kidney Foundation also offers a concise information page dedicated to the CKD-EPI equation [3]. The eGFR is expressed in milliliters per minute per 1.73 square meters ($mL/min/1.73m^{2}$), and the equation is as follows:

$$eGFR = 142*\min\left( \frac{standardized\ Scr}{K},\ 1 \right)^{a}*{\max\left( \frac{standardized\ Scr}{K},\ 1 \right)}^{- 1.200}*{0.9938}^{age}*1.012\ \lbrack if\ female\rbrack$$

Where:

• S_cr is the serum creatinine in mg/dL

• K is a constant, 0.7 for females or 0.9 for males

• a is a constant, -0.241 for females or -0.302 for males

• age is the patient’s age in years

The standardized S_cr

Standardized S_cr refers to laboratory measurements of serum creatinine (S_cr) that have been calibrated based on the National Institute of Standards and Technology (NIST) standard reference material (SRM) 967. References [4], [5] outline the methodology of the standardization and its importance in eGFR results, respectively. With the introduction of the SRM 967 in 2007, manufacturers have been required to follow this standard ever since [5].

The MIMIC-IV dataset

“MIMIC is a large, freely-available database comprising deidentified health-related data from patients who were admitted to the critical care units of the Beth Israel Deaconess Medical Center” [6]. While the database is freely-available, individuals are required to complete the necessary training in CITI Data or Specimens Only Research, and subsequently apply for data access through the physionet.org website. You can find my training certificate confirming completion of the required training here. Access to MIMIC is offered through two methods: cloud-based [7] or local [8]. For my master thesis I chose to extract the necessary data through BigQuery db. In this project, I have chosen to set up the local database, specifically utilizing a Postgres installation as outlined in the mimic-IV GitHub page, as I wanted to explore the capabilities of the PostgresSQL language.

Workflow

Here there is a list with the steps taken towards the dataset creation.

• Install the database locally. A comprehensive guide can be found here.

• Identify the tables needed for the specific project. Thankfully, the MIMIC project community provides some ready-made queries for creating concepts such as complete blood count or demographics. I have used the age query, the complete_blood_count and the chemistry from the demographics, and the diagnoses_icd from the hosp module. MIMIC comes with a comprehensive documentation found here.

• I created two sql queries and saved the results into two separate csv files.

  • The lab_query.sql extracts blood lab results (complete blood count and chemistry results) from the database for the same subject_id, with the same hadm_id where the blood samples have been taken simultaneously.

  • The patients_query.sql extracts basic patient demographics along with flags for diabetes, hypertension and ckd diagnoses for each patient (1 means positive diagnosis).

• I imported the csv files into a JupyterLab notebook where I joined the two Dataframes and proceded with further deidentification of the data using a SHA256 algorithm. I exported the results into a new csv file, and created this notebook, where the rest of the analysis takes place.

Issues and decisions

1. There are no urine lab results in the datasets.There is no derived query (like the chemistry query) that return lab results for urine tests. I have decided to not include anything from urine tests at the current version as it would bloat the project.

2. Since MIMIC-IV includes data from patient admissions between 2008 and 2019, I assume that S_cr laboratory measurements are standaridized.

3. The mimiciv_derived.age table contains a calculated age column where the values are floats containing many decimal places. I have decided to convert the column into int, in effect rounding the number, as in a hospital environment the whole part of the age is used most of the times.

Future versions

1. Create a query for urine lab results and modify the dataset.

2. Do a comprehensive statistical analysis of the dataset.

3. Check variations in eGFR calculation that might occur as a result of the age rounding decision.

References

[1] A. E. W. Johnson et al., ‘MIMIC-IV, a freely accessible electronic health record dataset’, Sci. Data, vol. 10, no. 1, p. 1, Jan. 2023, doi: 10.1038/s41597-022-01899-x.

[2] L. A. Inker et al., ‘New Creatinine- and Cystatin C–Based Equations to Estimate GFR without Race’, N. Engl. J. Med., vol. 385, no. 19, pp. 1737–1749, Nov. 2021, doi: 10.1056/NEJMoa2102953.

[3] ‘CKD-EPI Creatinine Equation (2021)’, National Kidney Foundation, Oct. 01, 2021. https://www.kidney.org/professionals/kdoqi/gfr_calculator/formula (accessed May 19, 2023).

[4] N. G. Dodder, S. S.-C. Tai, L. T. Sniegoski, N. F. Zhang, and M. J. Welch, ‘Certification of Creatinine in a Human Serum Reference Material by GC-MS and LC-MS’, Clin. Chem., vol. 53, no. 9, pp. 1694–1699, Sep. 2007, doi: 10.1373/clinchem.2007.090027.

[5] H. Pottel et al., ‘Standardization of serum creatinine is essential for accurate use of unbiased estimated GFR equations: evidence from three cohorts matched on renal function’, Clin. Kidney J., vol. 15, no. 12, pp. 2258–2265, Dec. 2022, doi: 10.1093/ckj/sfac182.

[6] ‘About MIMIC’, MIMIC. https://mimic.mit.edu/docs/about/ (accessed May 24, 2023).

[7] ‘Cloud’, MIMIC. https://mimic.mit.edu/docs/gettingstarted/cloud/ (accessed May 24, 2023).

[8] ‘Local database setup’, MIMIC. https://mimic.mit.edu/docs/gettingstarted/local/ (accessed May 24, 2023).

Footnotes

1. https://drive.google.com/file/d/1R_RTf9gSW9CxZMnw5xjPyk8NLDtpDk8B/view?usp=sharing ↩

2. https://github.com/fdiamant/Thesis ↩

3. https://www.kidney.org/sites/default/files/01-10-8374_2212_patflyer_egfr.pdf ↩

4. https://www.kidney.org/sites/default/files/kidney-numbers_ckd-heatmap.pdf ↩