Frequently Asked Questions

The most common KeySense questions we get asked about are - • Can I pause and continue a typing session? • The typing session is very long. Why can’t you make it shorter? • What is ‘dropoff’? • What do the other graphs mean? • My doctor has not taken the KeySense results seriously. What should I do? • What does the clinical diagnosis of Parkinson’s Disease by a neurologist involve? • How much do the KeySense results vary between sessions? • I have already been diagnosed with Parkinson’s, can I still do the assessment? • How accurate are the KeySense assessments? • How early in the disease can KeySense detect it? Here are the answers - Can I pause and continue a typing session? If you exit the page then your typing is not saved. However pausing for a few seconds (or even 10 minutes) is fine – you don’t have to be typing continuously. The typing session is very long. Why can’t you make it shorter? The session length is 2,000 characters. This is the shortest length we need for reliable calculation, so unfortunately it’s not possible to reduce this. What is ‘dropoff’? Dropoff is a measure of whether there is ‘fatigue’ in your finger movement during continuous segments of typing. Think of it like tapping your index finger against your thumb as fast as you can for 10 seconds. If you did that, could you keep up the repetition rate right to the end? Note that the KeySense dropoff measurement is not about typing speed itself – slow or one- fingered typing is fine. What do the other graphs mean? There are a total of 11 machine-learning models used within KeySense and we show the 4 most relevant of those. The labels are intended to describe broadly what each of them are doing – • ‘Inconsistency’ is based on statistical typing features, which can be summarised as unevenness in rhythm • ‘Dropoff’ is looking at slowing in repetition rate, as described in the previous question The combination of the 2 above detects bradykinesia and/or rigidity. Pauses in typing or hesitations in looking for a particular key won’t really affect the analysis. • ‘Sidedness’ looks at differences between your right hand and left hand features (someone with early PD typically has one side affected more than the other, but that is not a diagnostic criterion) • ‘Tremor’ uses machine learning to identify hand tremor in the range of 3.5 to 6 Hz, along with self-reported tremor (which you are asked about in the initial form). Unilateral tremor in that frequency range is suggestive of Parkinson’s Tremor My doctor has not taken the KeySense results seriously. What should I do? Doctors are generally quite conservative and are hesitant in embracing new tools. There is a basic problem with diagnosing Parkinson’s though - primary care doctors (GP’s and PCP’s) get it wrong 25% to 30% of the time. In contrast KeySense gets it wrong just 13% of the time, even with quite early symptoms. So if your KeySense scores are high, especially across multiple models, then we suggest that, if your doctor is dismissive, you consider getting a second opinion. Also keep in mind that KeySense can pick up motor symptoms 5 years or more before they may be visible (what is called ‘prodromal’), and a doctor not specifically trained in detecting Parkinson’s is very likely to miss those early cases. What are the criteria for a medical diagnosis of Parkinson’s? There are 4 cardinal features of Parkinson’s Disease - • Bradykinesia: Slowness of movement in the limbs, face, walking or overall body • Shaking or tremor: Called resting tremor, a trembling of a hand or foot that happens when the patient is at rest and typically stops when he or she is active or moving • Rigidity: Stiffness in the arms, legs or trunk • Posture instability: Trouble with balance and possible falls Until recently the clinical diagnostic criteria were the presence of bradykinesia, plus rest tremor or rigidity. However there are also a range of non-motor symptoms and these may be the thing that prompts the first visit to a doctor. Revised criteria from the International Parkinson and Movement Disorder Society have now come into use (MDS-PD) and this list reflects the most current understanding of the condition. It allows doctors to reach a more accurate diagnosis so patients can begin treatment at earlier stages, and includes a three-step process for PD diagnosis (see ‘MDS Clinical Diagnostic Criteria for Parkinson’s Disease’ and ‘Diagnostic Criteria for Parkinson’s Disease: From James Parkinson to the Concept of Prodromal Disease’) How much do the KeySense results vary between sessions? The answer to this is quite complex. Human circadian rhythms refer to the physical, mental, and behavioural changes that follow a 24-hour cycle. If you complete a second KeySense session at the same time of the day, then the classification (that is, ‘Unlikely’, ‘Slight possibility’ etc.) is similar in 90% of cases. However if it’s at a different time (e.g. afternoon compared to breakfast) then the results may be different. • In general, the KeySense scores will be higher in the early afternoon, but that isn’t always the case. The fluctuation (amount and timing) depends on many things, including your age, health status, medications, exercise, stimulants (such as coffee) and whether you are rested. • Avoid doing the assessment immediately after doing heavy labor involving the hands, as temporary muscle fatigue or pain may affect the results. • The actual scores of the individual AI models will vary somewhat between sessions. This is mainly to do with the limitation of having just 2/3 page of typing. So look at the overall classification rather than the raw scores. I have already been diagnosed with Parkinson’s, can I still do the assessment? Yes, but keep in mind that people already diagnosed are generally taking Parkinson’s medication of some type. So in that case KeySense will just be measuring the effectiveness of your medication at that particular time of day. How accurate are the KeySense assessments? Overall, KeySense achieves 95% accuracy in detecting Parkinson’s in its early stages. This has been validated against the corresponding diagnosis by neurologists. We also put the results into categories, and their respective accuracies are shown below: Score Range Category Level of confidence (for people over 55) 0 to .39 Unlikely Very high (typically >95% accuracy) .40 to .49 Slight possibility Moderate (around 80%) .50 to .59 Possible Moderate (around 80%) .60 to .69 Likely High (typically 90%) .70 or higher Very likely Very high (typically >95%) As you can see, the accuracy is highest for either low scores (less than 0.40) or high scores (more than 0.70). It is less certain for those in the middle ranges. • Keep in mind that even a 90% accuracy (such as the ’Likely’ category above) means that one in 10 people will get an incorrect result. • Note that the risk of Parkinson’s increases with age, and young-onset (for example, under age 55) is relatively rare. The KeySense results are also less accurate for younger ages (primarily due to limited ‘training’ data). How early in the disease can KeySense detect it? The KeySense methodology is very sensitive, which means it can detect symptoms of Parkinson’s quite early in the disease progression. In broad terms, the assessment categories can be equated with - Category Diagnosis Point Very likely Current diagnosis timeframe (by a general practitioner or non- specialist doctor), or several years earlier (i.e. you may have had some symptoms for 2 or 3 years). Likely 4 to 5 years earlier. That is, you may have had slight symptoms for up to 5 years. Non-specialist doctors may miss these early signs though. Possible or Slight Possibility Up to 6 or 7 years earlier, probably without visible motor symptoms, but perhaps with one or more ‘prodromal’ symptoms (which could include hyposmia, constipation, mood disorders and idiopathic REM sleep behavior disorder (RBD).

Contact Details

You are also welcome to contact us for any questions or feedback about KeySense. Please note that we cannot provide any type of medical advice. Our email address is admin@parkinsons-research.org