The CloudUPDRS smartphone software in Parkinson's study: cross-validation against blinded human raters npj Parkinson's Disease, 2020
The only study that runs 16 smartphone tests against the same blinded raters - 60 subjects, 990 tests, 2,628 blinded video MDS-UPDRS III ratings. Hand rotation predicted its clinical sub-item 74.6% of the time; finger tapping managed 53.2%. The authors note explicitly that this contradicts the field's habit of building tapping-only apps.
What Kampa did with it: shipped hand rotation alongside finger tapping rather than tapping alone. The same table scores a tremor test at 97% - we deliberately did not build it, because the Apple Watch already measures tremor passively and asking someone to perform their tremor on demand is worse data and a worse experience.
Read the paper →
Reliability and validity of the Roche PD Mobile Application for remote monitoring of early Parkinson's disease Scientific Reports, 2022
316 recently diagnosed, treatment-naive participants ran a short active-test battery at home every day. Adherence was 96.3%, the battery took a median of 5.3 minutes a day, and the pre-specified sensor features held up on retest - ICC at or above 0.75, median 0.9.
What Kampa did with it: this is the evidence that a short at-home check is something people will actually do, and that what comes out is stable enough to compare against itself over time. It is why a movement check exists at all. Kampa's version is seconds rather than a daily five-minute battery, and it is entirely optional.
What it does not show: ICC is test-retest reliability - the measure is consistent, not necessarily a measure of disease severity. Those are separate questions and we treat them separately.
Read the paper →
Sensor verification and analytical validation of algorithms to measure gait and balance and pronation/supination Sensors, 2022
Turn counts from the phone's gyroscope agreed with human raters counting turns by eye at ICC 0.935, and held at 0.97-0.99 even under deliberate perturbation - arm raised and lowered, stopping and starting, rotating about several axes.
What Kampa did with it: adopted the protocol - phone held in the hand, arm out, screen palm-up, rotate palm-up to palm-down - and the gyroscope-plus-PCA approach that finds the true axis of rotation, which is what makes the measurement survive someone who doesn't hold a clean axis at home.
What it does not show: this validation was run in healthy volunteers. It establishes that the algorithm measures rotation correctly, not that the measurement tracks Parkinson's severity - that is what the two studies above are for.
Read the paper →
Quantification of limb bradykinesia in patients with Parkinson's disease using a gyrosensor International Journal of Precision Engineering and Manufacturing, 2011
Rotation measured by gyroscope correlated at r = -0.78 with the hand-rotation bradykinesia sub-score, and correlated better with that sub-score than with others.
What Kampa did with it: treated rotation as a measure of the specific thing it claims to measure, rather than a proxy for general severity.
Read the paper →
Smartphone-based estimation of item 3.8 of the MDS-UPDRS-III for assessing leg agility IEEE Open Journal of Engineering in Medicine and Biology, 2020
A phone strapped to the thigh scored r = 0.92 against clinicians - higher than the clinicians' agreement with each other, at 0.88. Held against the thigh by hand instead, accuracy drops to the 63-70% range.
What Kampa did with it: nothing yet, on purpose. The accurate version needs a strap we can't ship, and the hand-held version puts hand tremor into a leg measurement - a problem the literature doesn't address. The test is sequenced behind rotation rather than shipped weak.
Read the paper →