The Dirty Nappy Study

The Dirty Nappy Study: Development and Evaluation of a Machine Learning Algorithm for the Early Detection of Cholestasis and Biliary Atresia From Parent-Provided Images of Dirty Nappies

This observational study evaluates whether a machine-learning algorithm, a computer program that learns patterns from data, can accurately diagnose cholestasis in newborns. Cholestasis refers to reduced or blocked bile flow from the liver, which can lead to liver damage. A severe form of cholestasis is biliary atresia, a condition where the bile ducts are damaged or absent, requiring early treatment to prevent long-term harm.

The study involves infants from birth, both healthy and those potentially affected by cholestasis, recruited from four UK hospitals. It addresses two primary aims:

• Accuracy of Diagnosis: Can the machine-learning algorithm accurately identify cholestasis and biliary atresia using parent-provided stool images? This will be assessed by measuring sensitivity (the ability to correctly detect true cases) and specificity (the ability to correctly identify infants without the condition).
• Feasibility of Screening: Is using parent-provided images a feasible and acceptable screening method for early detection?

To evaluate these aims, researchers will compare two groups:

• Infants with abnormal stool images who are subsequently diagnosed with cholestasis or biliary atresia.
• Infants with normal stool images who do not develop biliary atresia.

This comparison will help determine the algorithm's ability to distinguish between infants with and without these conditions.

Parents will:

• Take smartphone photos of their baby's dirty diapers at 14, 21, and 28 days of age.
• Upload the images for analysis by the algorithm.
• Provide feedback on their experience with this screening process.

The study seeks to determine if parent-submitted stool images can serve as a practical early screening tool for cholestasis, potentially enabling faster diagnosis and improved outcomes for affected infants.

Study Overview

• Status: Recruiting
• Conditions: Biliary Atresia; Cholestasis in Newborn; Neonatal Cholestasis

Detailed Description

As the Study Design section only allows selection of a single time perspective, this Detailed Description box is being used to clarify that the study involves both prospective and retrospective data collection, as instructed.

This study includes two prospective arms (following newborns from birth, with and without suspected cholestasis, and collecting stool images in real time) and one retrospective arm (using historical stool images and clinical data from diagnosed cases).

• Study Type: Observational
• Enrollment (Estimated): 5350

Contacts and Locations

Study Locations

• United Kingdom
  • Birmingham, United Kingdom
    • Recruiting
    • Birmingham Women's and Children's NHS Foundation Trust
    • Principal Investigator: Girish Gupte
    • Contact: Girish Gupte; Phone Number: +44 121 333 9999; Email: girishgupte@nhs.net
    • Contact: Suzanne Peters; Email: suzanne.peters2@nhs.net

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

Recruited from hospitals enrolled in the study.

Description

Inclusion Criteria:

All Arms:

Informed Consent: Parents/guardians can give informed consent for their child's participation in the study and can understand written English.

1. Retrospective Cholestasis Arm:

   All infants, of any gestation, male and female who were referred to Birmingham Liver Unit from any hospital for investigation of suspected cholestasis within the previous 10 years before the study launch. At the time of the referral to Birmingham Liver Unit the children were under the age of 6 months and the children were referred before the start date of the study.

2. Prospective Cholestasis Arm:

   All infants, of any gestation, male and female who were referred to Birmingham Liver Unit from any hospital for investigation of suspected cholestasis after the study launch. At the time of the referral to Birmingham Liver Unit the children were under the age of 6 months.

3. Prospective Non-Cholestasis Arm All infants, of any gestation, male and female & aged less than 28 days, not thought to be cholestatic at the time of recruitment & with the birth registered at any of the participating hospital trusts.

Exclusion Criteria:

If the inclusion criteria as outlined are met, there is no exclusion criteria.

Study Plan

How is the study designed?

Number of groups / cohorts

3

Cohorts and Interventions

Group / Cohort
Retrospective Cholestasis Cohort; This group includes infants who were referred to a specialist paediatric liver unit for suspected cholestasis within the past 10 years. This cohort provides historical data to evaluate the machine-learning algorithm's performance using existing images and clinical outcomes. No new interventions or prospective data collection occur in this group.
Prospective Cholestasis Cohort; This group includes infants who are newly referred for cholestasis investigation to a specialist paediatric liver unit during the study period. Parents will provide stool images at 14, 21, and 28 days of life, 3 months, and 6 months using a smartphone. These images will be analysed by a machine-learning algorithm to assess whether early stool image screening can help detect cholestasis, including biliary atresia. Participants will also share feedback on their experience with this screening method. This cohort provides real-time prospective data to train and validate the algorithm.
Prospective Non-Cholestasis Cohort; This group consists of newborns born at specified hospitals who are the time of recruitment are not thought to be cholestatic. Parents will submit stool images at 14, 21, and 28 days of life, 3 months, and 6 months to provide a large dataset of normal stool images. This cohort serves as a control group, ensuring that the algorithm can distinguish between healthy and abnormal stool patterns.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity & Specificity of an Algorithm in Detecting Cholestasis	Sensitivity: The proportion of stool images from clinically confirmed cholestatic infants that are correctly classified as positive by the algorithm, calculated as a single estimate across all evaluable images collected between Day 14 and Day 34 of life, using an independent held-out test dataset. Specificity: The proportion of stool images from clinically confirmed non-cholestatic infants that are correctly classified as negative by the algorithm, calculated as a single estimate across all evaluable images collected between Day 14 and Day 34 of life, using an independent held-out test dataset.	Stool images collected between Day 14 and Day 34 of life; evaluation performed once clinical confirmation of cholestasis status is complete (up to 6 months of age).

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity & Specificity of an Algorithm in Detecting Biliary Atresia	Sensitivity: The proportion of stool images from clinically confirmed infants with biliary atresia that are correctly classified as positive by the algorithm, calculated as a single estimate across all evaluable images collected between Day 14 and Day 34 of life, using an independent held-out test dataset. Specificity: The proportion of stool images from clinically confirmed infants without biliary atresia that are correctly classified as negative by the algorithm, calculated as a single estimate across all evaluable images collected between Day 14 and Day 34 of life, using an independent held-out test dataset.	Stool images collected between Day 14 and Day 34 of life; evaluation performed once clinical confirmation of biliary atresia status is complete (up to 6 months of age).
Sensitivity & Specificity of an Algorithm in Detecting Abnormal Stool Appearance (Colour and/or Texture)	Sensitivity: This measure is the proportion of abnormal stool images from infants that are correctly classified as positive by the algorithm. Sensitivity will be calculated separately for images collected during the third, fourth, and fifth weeks of life, yielding three distinct sensitivity values. This approach provides an independent assessment of how well the algorithm identifies true positive results, using a separate, untrained test dataset. Specificity: This measure is the proportion of normal stool images from infants that are correctly classified as negative by the algorithm. Specificity will be calculated separately for images collected during the third, fourth, and fifth weeks of life, yielding three distinct specificity values. This approach provides an independent assessment of how well the algorithm avoids false positive results, using a separate, untrained test dataset.	Images collected in the third week (Day 14- Day 20), fourth week (Day 21- Day 27), and fifth week (Day 28- Day 34) of life; evaluation is performed once clinical confirmation is complete (up to 6 months of age).
Sensitivity & Specificity of an Algorithm in Detecting Biliary Atresia in Cholestatic Infants	Diagnostic accuracy of the algorithm in detecting biliary atresia among cholestatic infants, reported as sensitivity and specificity. Both measures share a common unit of measure (proportion, reported as percentage). Sensitivity: the proportion of cholestatic infants with confirmed biliary atresia whose combined stool image and supplemental clinical data are correctly classified as positive by the algorithm. Specificity: the proportion of cholestatic infants without biliary atresia whose combined stool image and supplemental clinical data are correctly classified as negative by the algorithm. Both assessed using an independent held-out test dataset not used during model training.	Data collected from children less than 6 months; evaluation is performed once clinical confirmation is complete (up to 6 months of age).
Impact of Repeat Stool Image on Diagnostic Reclassification Among Initially Positive Cases in the 3rd Week of Life	Among infants initially classified as positive for cholestasis / biliary atresia by the Elixir algorithm based on stool images submitted during the 3rd week of life (Day 14-20), this outcome measure evaluates the change in diagnostic classification after a repeat stool image is submitted during the subsequent window (Day 21-27). Specifically, it will assess: The proportion of initially positive cases that are reclassified as negative upon repeat testing, and are subsequently confirmed as false positives.; The proportion of initially positive cases that remain positive upon repeat testing and are confirmed as true positives. This measure is intended to determine whether requesting a repeat stool image can reduce the false positive rate without inadvertently reclassifying true cholestatic cases as negative (i.e. false reassurance). Final classification for each infant will be confirmed by clinical diagnosis up to 6 months of age.	Repeat stool images will be collected between Day 21 and Day 27 of life for infants with an initial positive result from Day 14-20; final outcome confirmation will occur through clinical follow-up (up to 6 months of age).
Parents / Guardians Comfort Level with Photo-Based Diagnostic Requests to Assist in the Diagnosis of Biliary Atresia	Parents' comfort levels with receiving and responding to requests via text message to provide photographs of their child's dirty nappies to assist in diagnosing biliary atresia. Comfort is assessed quantitatively using a standardised five-point Likert scale ranging from "Very comfortable" to "Very uncomfortable." Qualitative follow-up captures additional explanatory context for ratings and will be analysed separately.	Assessments will be completed by parents/guardians at the end of each child's participation in the study (up to 6 months of age).
Parents / Guardians Comfort Level with Smartphone-Based Health Information Sharing, Including Photos, Videos, Audio, or Questionnaire Responses, to Assist in Medical Diagnosis or Prioritise Care	Parents' comfort levels regarding sharing child health-related information with medical professionals using smartphone technology (including photos, videos, audio, or questionnaire responses) to assist in medical diagnosis or prioritise care. Comfort is assessed quantitatively using a standardized five-point Likert scale ranging from "Very comfortable" to "Very uncomfortable." Qualitative follow-up captures additional explanatory context for ratings and will be analysed separately.	Assessments will be completed by parents/guardians at the end of each child's participation in the study (up to 6 months of age).
Sensitivity & Specificity of the Current UK Practice of Prolonged Jaundice Screening in Detecting Cholestasis	The UK practice of prolonged jaundice screening, is an assessment on whether or not to send the child for a split bilirubin blood test, performed by a health visitor or midwife after day 14 of life for term infants and after 21 days of life if premature. Sensitivity: This measure is the proportion of infants clinically confirmed with cholestasis who are correctly identified by the existing UK prolonged jaundice screening. Specificity: This measure is the proportion of infants without cholestasis who are correctly identified as negative by the same screening. Data will be derived retrospectively from anonymised hospital referral records. Results will be compared with sensitivity and specificity metrics from the proposed diagnostic algorithm evaluated elsewhere in this study.	Data collected retrospectively from historical records covering a one-year period.
Sensitivity & Specificity of the Current UK Practice of Prolonged Jaundice Screening in Detecting Biliary Atresia	The UK practice of prolonged jaundice screening, is an assessment on whether or not to send the child for a split bilirubin blood test, performed by a health visitor or midwife after day 14 of life for term infants and after 21 days of life if premature. Sensitivity: This measure is the proportion of infants clinically confirmed with biliary atresia who are correctly identified by the existing UK prolonged jaundice screening. Specificity: This measure is the proportion of infants without biliary atresia who are correctly identified as negative by the same screening. Data will be derived retrospectively from anonymised hospital referral records. Results will be compared with sensitivity and specificity metrics from the proposed diagnostic algorithm evaluated elsewhere in this study.	Data collected retrospectively from historical records covering a one-year period; final outcome confirmation will occur through clinical follow-up (up to 6 months of age).
Mean Age at Time of Split Bilirubin Testing Following the Current UK Practice of Prolonged Jaundice Screening	This measure captures the mean age (in days) at which infants undergo a split bilirubin blood test, following referral to hospital through the current UK prolonged jaundice screening, an assessment on whether or not to send the child for a split bilirubin blood test, performed by a health visitor or midwife after day 14 of life for term infants and after 21 days of life for premature infants. Data analysis will include mean age, range, standard deviation, and inter-hospital variability. Anonymised clinical records from participating hospitals will be retrospectively analysed. These results will be compared with the age of split bilirubin projected by the proposed diagnostic algorithm evaluated elsewhere in this study.	Data collected retrospectively from historical clinical records covering a one-year period.
Mean Age at Time of Hepatoportoenterostomy (HPE) Surgery Following the Current UK Practice of Prolonged Jaundice Screening	This measure assesses the mean age at the time of surgical intervention (HPE) following clinical diagnosis of biliary atresia the current UK prolonged jaundice screening, an assessment on whether or not to send the child for a split bilirubin blood test, performed by a health visitor or midwife after day 14 of life for term infants and after 21 days of life for premature infants. Reported as days from birth until surgery. Data analysis will include mean age, range and standard deviation. Data will be obtained retrospectively from anonymised surgical records and clinical notes. These results will be compared with the projected timing of intervention enabled by the proposed diagnostic algorithm evaluated elsewhere in this study.	Data collected retrospectively from historical surgical records covering a one-year period.

Collaborators and Investigators

• Sponsor: Birmingham Women's and Children's NHS Foundation Trust
• Collaborators: Elixir AI

Study record dates

Study Major Dates

• Study Start (Actual): April 1, 2025
• Primary Completion (Estimated): September 30, 2026
• Study Completion (Estimated): September 30, 2026

Study Registration Dates

• First Submitted: February 28, 2025
• First Submitted That Met QC Criteria: July 6, 2026
• First Posted (Actual): July 13, 2026

Study Record Updates

• Last Update Posted (Actual): July 13, 2026
• Last Update Submitted That Met QC Criteria: July 6, 2026
• Last Verified: March 1, 2026

More Information

Terms related to this study

• Keywords: machine learning; screening; AI; image analysis; Neonatal cholestasis; Biliary atresia diagnosis; Newborn stool analysis; infant liver disease
• Additional Relevant MeSH Terms: Digestive System Diseases; Biliary Tract Diseases; Congenital Abnormalities; Bile Duct Diseases; Digestive System Abnormalities; Congenital, Hereditary, and Neonatal Diseases and Abnormalities; Biliary Atresia
• Other Study ID Numbers: 327150

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No