It is 19:15 and daylight savings time.   It is too hot to go out and run or bike.  Time to write.

Today we were helping a customer with hardware issues. At the end of a long day, I started thinking that even hardware issues are valuable data to the decision-making process of measuring efficacy of treatment.

We specialise in reducing time to regulatory submission in clinical trials.  Our medtech customers use the Flaskdata.io platform to collect data from patients, devices and investigators and automate and prove efficacy of their device.  As Skolnick et al note in Compliance, Compliance, Compliance – the secrets of a successful clinical trial :

Although more commonly considered to be a phase 1 phenomenon, efficacy trials also attract professional subjects, particularly when entry criteria and endpoints are “soft,” such as trials using subjective rating scales which can be “gamed.”

Not good.  How do you measure a valid clinical response? The VAS scale for pain reduction can be gamed so is it a simple measure of pain reduction sufficient? My thesis advisor in solid state physics always said that he only does simple things (simple may be a mantra that many grad students hear while they trudging on a steep uphill slope of research).

Simple is not always sufficient

When you rely on a simple mode of clinical response you get a partial picture.

The clinical response to treatment is an important indicator of the therapeutic effect of a medtech device or drug. The value and interpretation of clinical response has to be carefully considered within the intended use by the patient.

Let’s consider a simple model of clinical response that is based on cause and effect.

-You use an electroceutical device and the pain is reduced as measured by a VAS scale.

-You stimulate vagus nerve and the patient has a bowel movement as measured by a ePRO.

Is this sufficient?  Maybe maybe not.   Did the device perform as expected?   Did the patient game the treatment?

A better model of assessing clinical response

Response assessment should be combined with other indicators of the patient’s condition to contribute to the decision-making process.

We can use additional connected devices in order to measure biological response and biometrics. With additional data, we can validate patient compliance and proper operation of the device.  We can also detect and respond to patients who game the treatment.

One example that comes to mind is measuring peptide levels in saliva (which requires a lab test) or skin temperature (which can be measured directly). Another example is weight differences between medication-compliant cohort and the placebo group. Weight loss may be indication of medication adherence in certain treatments and it is simple to measure.

For home-use devices, we can consider proper operation of the device itself. The device should provide 3 kinds of introspection.  (Introspection means that the device tells us how it’s doing and how it was used)

1.The device should record whether or not it functioned properly when used. A simple success/fail code will do.
2.The device should record proper operation; whether or not it was used properly. A simple count of number of operations will be helpful.
3.The device should record timestamps of operation to enable comparison with timestamps in the device log. The best solution is for the device to transmit log records via an API to a secure service with its internal timestamps. The timestamps will tell us if the device is using NTS properly, or if the clock is drifting. The timestamps will also help analyze multiple activations (mistakes or professional patients gaming the treatment).

More data helps us detect and respond to patient and device compliance issues

Device introspection will reveal that the patient misused the device, or the device failed. The first case is an issue of patient compliance. The second case is an issue of the device not providing the treatment as designed.

Listening to the device and to the body

The model of clinical response to treatment can be enriched with additional data sources from the device itself as well as biological response and biometrics measurements.

A richer model helps us detect and respond to patient compliance deviations.   The additional data also helps us understand if the investigational product is defective.   This is an advantage that medtech has over drugs.