AIMxHIE discussion - Taimaka: Improving MUAC (Mid-Upper Arm Circumference) measurement (3rd March 2026)

[aaronboddy]

Taimaka is a global health charity, working to scale a cost-effective treatment model for addressing pedriatric malnutrition.

On 3rd March, their co-founder Justin Graham will join us to tap into the wisdom of the High Impact Engineers community to improve the measurement solution for measuring a child's mid-upper arm circumference (MUAC).

Sign up link is here: https://luma.com/i0b6l78d

Here's some more context from Justin:

The core part of malnutrition treatment is measuring a child's mid-upper arm circumference (MUAC). It is what is used to diagnose children with malnutrition, track their progress, and declare them healthy at the end of treatment. Currently, MUAC is measured by using a MUAC tape (just a colored armband that wraps around someone's arm) - see here.
This methodology is not great for two reasons:

1. What reading you get depends on how tightly you pull the band, and everyone will pull a band slightly differently, so getting similar readings between different staff members is hard. This can lead children to get discharged when they aren't yet healthy, because they get mismeasured. It also makes it hard to enforce uniform standards, because measurements vary so widely, it is hard for people to check each other's work. This reduces patient care quality.
2. In part because of the above, and in part because all we have to verify a measurement's value, it is easy to fake a MUAC value, to make a child appear malnourished or recovered when they are not. Someone might do this for a few reasons, like trying to get a non-malnourished child who is the child of a friend of theirs admitted so they get access the program (wasting resources), or faking a child being recovered to avoid criticism of poor work (hurting that child).

• i) This means we have to spend a lot of time and effort backchecking people's values (which is hard, again because of 1), and makes it so that we have to have our own, high-trust staff measure each child.

• ii) We'd like to experiment with gig-based malnutrition work by community health workers where they get paid per visit to a malnourished child, but we can't do that without a more verifiable measurement system, as otherwise they have a clear incentive to lie about whether a child is malnourished so they can get paid more.

These are related problems that might have different solutions. E.g., 1 might be solvable with a constant-tension band of some kind, or a cuff that won't fit children with a MUAC of >= 12.5 cm, which is the cutoff point for being healthy. 2 is trickier because it requires verifiability without a supervisor being present, and you can always fudge observations with physical tools. Some sort of ML model that can identify arm circumference, or use of things like an iPhone's LIDAR sensors to do measurement, are things that I have thought of so far.

It would be cool if people could also start the conversation in the Impact Forge, so that we can really hit the ground running on the call itself :)

[gwfellows]

tape measures are approximately constant-force springs and small ones are very cheap. You could make a simple frame for a tape measure that attaches to the end of the tape, creating a loop. Then you just pull the loop open, insert the patient's arm, let the tape measure retract to close the loop, and read the value.

still does not fix problem 2

[JoePaterJoePaterJoePater]

Love that idea! Very clever.

Problem 2 seems very difficult - even if your equipment is tamper proof, how can you verify that you're actually measuring the right child? Maybe some sort of photographic smartphone app?

[rhjiang]

I think the tape measure idea is really neat! A few additional thoughts...

• If the tape measure coil is too cumbersome, you could do an in-line spring with a ruler behind it, and technicians are instructed to pull until the spring has extended by some length (indicating some force target has been met).
• The simplest way I can think to make this SLIGHTLY harder to lie about the measurement is to make the band automatically digitally log the measurement when the spring force target is met. However, it's still pretty easy to cheat with this. e.g., over-report circumference by putting a finger in the band loop alongside the arm, or under-report circumference by using the wrong part of the arm. Though, some basic level of employee supervision should prevent these things (whereas they wouldn't have prevented someone from pulling a bit too hard on the band or leaving it a bit too loose).
• If we want to go more expensive/possibly tech overkill, we could think about things like:
  • A cuff that has lidar sensors pointing inward to map the surface of the arm + laser indicators so technicians can see that they are measuring the right slice of the arm
  • Scanning the entire upper body somehow, to verify that it's the right child alongside the arm measurements
    • Could think about using something like neural radiance fields
  • Some existing clinical devices like https://medirent.com.au/rental-products/perometer-led-scanner
  • Throw learning at it -- e.g., take a phone video that captures the face and arm from a few angles. Face is used to ensure it's the right child. Arm is used to estimate arm circumference. Perhaps you also still use the band, and if the two estimates differ by more than some threshold, you send the case for someone higher up to verify or you re-take the band measurement and scan.

[amanda-matthes]

Super cool initiative!

It sounds to me like you mostly have a game theory problem (on top of maybe a hardware problem).

An idea that might be worth discussing: Instead of using one band that measures in centimetres (or some other known, repeated unit), use e.g. 50 different bands that all have a different, somewhat arbitrarily chosen unit. Give every person doing measurements a different band (and maybe rotate). Then, instead of logging a circumference as e.g. 20cm, they log something like: 15 on band number 7. Since you made the bands, you can convert that to an actually useful unit. That way, you can then catch when repeated measurements don't add up.

People could still present the wrong child for measurements or treatments. So you might still want to use a photo or some other ID method.

Do you have data on how much good faith measurements actually vary?

[aaronboddy]

Thanks to everyone who attended the call on 3rd March!

Here are the notes from my AI Notetaker for the benefit of those who weren't able to attend (and as a reminder for those who were!)

Feel free to continue the conversation in this channel.
And we might schedule a follow-up call at some point in the future :)

Problem Overview

• Taimaka provides cost-effective acute malnutrition treatment in northeastern Nigeria, treating severely malnourished children
• Current diagnostic tool is MUAC (mid-upper arm circumference) tape - a color-coded tape measure used to assess child malnutrition status
• Key diagnostic thresholds: moderate acute malnutrition at 12.5cm, severe acute malnutrition at 11.5cm
• Current measurement precision is ±0.2-0.4cm even with skilled measurers, making it difficult to track weekly progress when children's MUAC only changes by 0.1-0.2cm per week
• Problem is magnified when measurements are taken by less-skilled community mobilizers rather than trained anthropometrists

Why Measurement Precision Matters

• Tracking recovery trajectory: Children visit facilities 6-10 weeks for treatment, need to track if MUAC is increasing week-over-week to identify non-responders early
• Impact verification: Need to verify children are genuinely malnourished at admission and recovered at discharge, but current imprecision makes this difficult
• Preventing fraud: Incentives exist to fake malnutrition (to obtain therapeutic food) or inflate recovery rates, making accurate measurement critical

Current Measurement Challenges

• Inconsistent tension: Everyone pulls the tape to different tightness levels, creating variation between measurers
• Midpoint identification: Finding the mid-upper arm requires additional steps and introduces error
• Friction issues: Tape friction through measurement holes causes measurements to stick at higher values rather than relaxing to correct tension
• Training requirements: Achieving excellent measurement precision (±0.27-0.33cm) requires extensive training, practice, and quarterly recertification

Gig-Based Worker Model Context

• Experimental model would use app-based gig workers instead of facility-based staff
• Workers would identify, treat, and track malnourished children in their communities, paid per visit
• Benefits: more operational flexibility, easier to scale geographically, can adjust coverage areas without opening/closing facilities
• Challenge: increased fraud risk with financial incentives to find/treat malnourished children
• Payment structure could include finder's fee, small weekly payments, and larger lump sum upon recovery to align incentives
• Would require smartphone-based verification solution with ~85% accuracy threshold

Fraud Prevention Through Information Control

• Alternative approach: provide measurement bands without standard units (centimeters/colors) - each band has unique number system
• Makes it difficult for individuals to game measurements, allows statistical detection of outliers
• Helps at facility level where multiple people measure same child week-to-week
• Less applicable to community gig model where typically only one person sees child
• Would require probabilistic model to distinguish measurement imprecision from actual changes over time

Background on MUAC Measurement

• MUAC historically mapped to weight-for-height z-scores (more than 3 standard deviations below normal = severely malnourished)
• WHO funded multi-country study to create mapping between MUAC and weight-for-height z-scores
• MUAC more correlated with child mortality than weight-for-height z-score, despite being constructed index
• Mortality probability is ultimate metric of interest - malnutrition diagnosis is proxy
• Treatment program cost is approximately $100 per child, with $35-40 being therapeutic food for 6-10 weeks

Hardware Solutions Discussed

Auto-Retracting Tape Measure (Grant's Solution)

• Use standard flat-band tape measure with end taped back to create self-tightening loop
• Spring provides roughly constant tension, eliminating variable tightness problem
• Simple, inexpensive solution that could be deployed immediately
• Concern about spring tension changing over time with repeated use

Elastic Band Solution

• Attach elastic string to tape end, instruct operators to stretch to specific length (e.g., 2cm extension)
• Provides controlled, measurable tension proportional to elastic stretch
• Potentially cheaper and more appropriately sized than spring scales
• Testing showed comfortable tension around 2 newtons

Gulick Tape Adaptation

• Gulick 2 tape includes spring force measure with red line indicator for consistent tension
• Previous trial found it required too many hands while managing squirming children
• Friction between tape and attachment point remains an issue

Material Improvements

• Consider using slipperier tape material to reduce friction
• Try wider tape positioned at top of arm under armpit to eliminate midpoint-finding step

Camera-Based Measurement Solution

Core Concept

• Use smartphone camera to photograph child's arm with reference sticker of known size
• Assume arm is cylindrical, calculate circumference from width measurement
• Could work for both facility-based and community gig worker models

Implementation Approach

• Place circular sticker of known size on child's mid-upper arm
• Take standardized photo with child's arm positioned consistently (e.g., arm at side, palm backward)
• Use sticker as reference to calculate arm width, then circumference
• Consider April tag (advanced QR code) for precise orientation tracking

Advantages

• Provides both measurement and child identification through facial recognition
• Creates verifiable record that can be audited later
• Portable solution that works at facility and home level
• Leverages existing smartphone infrastructure already used for digital care guides

Accuracy Considerations

• Need to validate accuracy through correlation with observed MUAC measurements
• Requires controlled photo orientation and standardized pose
• Simple version: child against wall, camera at fixed height, circular sticker
• Advanced version: April tag with app that auto-captures when orientation is correct
• Could use cardboard box rig with built-in scale backdrop for consistent positioning
• Expected precision likely around 0.4cm or better

Cross-Sectional Shape Considerations

• Accuracy depends on how circular malnourished children's arms actually are
• Need to collect data on actual arm cross-sections to validate cylindrical assumption
• Orientation matters less if standardized pose is enforced

[rhjiang]

I'd be curious to hear if anything comes of this!