Predictive health
Predictive health seeks information about disease risk before illness becomes apparent, with the aim of maintaining health and enabling earlier action. Its usefulness depends on obtaining interpretable data, developing predictive rules and understanding what a prediction cannot establish.
Purpose and proposed benefits
Predictive health has been proposed as finding the causes of disease and intervening before it occurs, shifting attention from treating established disease towards maintaining health. Its aim is to acquire specific information that can identify an individual or group at risk of a particular disease, with individualised treatment a further possibility.
Four proposed benefits are:
- Tools for defining health positively.
- Information that helps maintain health.
- Integration of research and healthcare to move discoveries into application.
- Education of biomedical professionals with roles different from traditional healthcare roles.
Computational biology, molecular imaging, stem-cell research and high-throughput genomics, metabolomics and proteomics offer ways to investigate biology and measure health.
Data and predictive rules
Quantifying health and disease involves several related activities.
- Comprehensive research.
- Targeted biological research examining a particular gene in depth.
- Exploratory research comparing genes upregulated or downregulated under the same conditions.
- Data management and bioinformatics.
- Computational systems modelling.
Developing health predictions requires obtaining data, extracting information and analysing both with appropriate methods. Acquisition, storage and access present technical, logistical and ethical challenges. Useful predictive rules require integrating knowledge and tools from biology, chemistry, physics, mathematics, engineering and computing.
Repeated and behavioural measurements
Episodic clinical visits leave gaps between observations, whereas repeated measurements can capture changes over time. In 2018, a Verily representative reported that its Baseline study collected extensive biomedical measurements from consenting volunteers, with the first two days potentially generating six terabytes if every test were run on every collected sample. Participants also used a phone, watch and sleep sensor; engineering and human-factor difficulties included distinguishing the participant’s sleep from effects of pets sharing the bed.
Digital phenotyping explores markers in how people interact with digital devices, such as tremor during phone use, keyboard interaction associated with cognitive dysfunction, or a reduced radius of activity associated with depression. Household and geographical information can also affect health-risk prediction because people share habits and exposures.
Limits of genetic prediction
Genetic results can mislead through false positives, including an incorrectly detected variant or an incorrect association between a variant and disease. Detected variants may have unknown or uncertain significance, and even disease-associated variants may not be actionable. False negatives can provide inappropriate reassurance and delay medical consultation, losing an opportunity for earlier treatment.