Predictive and Preventative Care: Using Data to Keep People Well at Home

From reactive to proactive care

For most of its history, domiciliary care has been organised around delivery rather than anticipation. A carer arrives, provides the commissioned support, records what happened, and leaves. When something goes wrong (a fall, an infection, a sudden decline in a long-term condition), the response is typically escalation: a call to the GP, an out-of-hours service, or, most expensively of all, an emergency admission. The intelligence needed to intervene earlier is often present, but scattered across handwritten notes, individual carers' memories and separate NHS systems that rarely talk to one another.

The alternative that policymakers increasingly describe is a preventative model: one that segments a population by need, identifies who is most likely to deteriorate, and acts before the crisis. NHS England defines population health management as exactly this: a data-driven, proactive approach that uses risk stratification to find at-risk cohorts and intervenes early to “keep people well” rather than waiting to respond to emergencies (NHS England, 2025). Whether that ambition can be realised in home care specifically is the question this essay tries to answer honestly.

The clinical and economic case for prevention

The demographic pressure behind the shift is not subtle. In England, healthy life expectancy at birth in 2021-2023 was roughly 61.5 years for males and 61.9 years for females, and it has fallen since 2017-2019 (ONS, 2024). Set against total life expectancy, that figure implies close to two decades of later life lived in less than good health, years in which people are managing multiple long-term conditions, frailty and rising dependency, very often at home. The task facing home care is therefore not simply to expand, but to compress that period of poor health where it can.

The workforce arithmetic sharpens the point. Skills for Care projects that the adult social care sector in England will need in the order of 470,000 additional posts (a rise of around 27%) by 2040 to keep pace with a growing over-65 population, and it must do so against persistent vacancies and turnover (Skills for Care, 2024). If demand cannot be met simply by adding staff, then any intervention that keeps people stable at home, and out of the acute system, has both a human and an economic logic. Prevention, in this framing, is not a wellness slogan; it is a capacity strategy for a sector that cannot recruit its way out of the demographic curve. We explore that staffing dimension further in our essay on the future of the care workforce.

Risk stratification: predicting decline and falls

If prevention depends on knowing who is most at risk, the obvious move is to build predictive tools from data already collected. The most mature example in UK practice is the electronic frailty index (eFI), constructed from 36 cumulative deficits in routine primary-care records across roughly a million patients; it predicts mortality, unplanned hospitalisation and care-home admission with reasonable discrimination (Clegg et al., 2016). Tools of this kind now sit inside GP systems and increasingly inform how integrated care systems target support.

The temptation is to assume that a good predictive model automatically produces better outcomes. It does not. The PRISMATIC trial, a randomised, stepped-wedge study of a validated predictive risk-stratification tool (PRISM) introduced into Welsh general practice, found that its arrival was followed by more emergency admissions, more A&E attendances and higher cost, with no clear patient benefit (Snooks et al., 2018). The most likely explanation is supply-induced demand: flagging a patient as high-risk prompts activity, and activity is not always in the patient's interest. A score that identifies risk is not the same as a plan that reduces it.

Falls prediction illustrates the limit even more starkly. Falls are a leading cause of injury, admission and loss of independence in older people, so a reliable falls-risk predictor would be enormously valuable. Yet the 2025 NICE guideline is unambiguous: it recommends against using falls-risk prediction tools to estimate an individual's risk of falling, because their sensitivity and specificity are too poor and they merely sort people into tiers without guiding what to do next (NICE, 2025). The instructive lesson for home care is that stratification is a beginning, not an answer. Its value depends entirely on whether a proportionate, evidence-based intervention follows.

Remote monitoring, wearables and sensors

Alongside record-based prediction sits a second strand: continuous or near-continuous monitoring in the home, through telecare alarms, motion and door sensors, and a growing range of wearables. The evidence here is genuinely encouraging but also cautionary. The Whole System Demonstrator trial, the largest UK randomised study of home telehealth, found that telehealth was associated with lower twelve-month mortality (4.6% versus 8.3%) and fewer emergency admissions (odds ratio 0.82) (Steventon et al., 2012). Those are meaningful signals. However, the companion cost-effectiveness analyses were disappointing, which points to a crucial condition: monitoring helps only when it is embedded in a service that can respond to what the data shows. A sensor that alerts no one, or alerts a service with no capacity to act, changes nothing.

There is also a large, unglamorous infrastructure problem beneath the whole enterprise. The migration of the UK telephone network from analogue to digital threatens the very telecare estate that an estimated 1.7 million users depend on; documented device and call-failure risks were serious enough that the switch-off deadline has been pushed back towards 2027 (Hamblin & Whitfield, 2025). Preventative monitoring, in other words, rests on plumbing that is itself in flux, a reminder that technological ambition and operational reality are not the same thing. We examine that transition in detail in our piece on technology-enabled care and the analogue switch-off.

Admission avoidance and the quiet power of continuity

The clearest destination of a preventative model is fewer avoidable hospital admissions. Here the evidence is careful rather than triumphant. A Cochrane review of admission-avoidance “hospital at home” concluded that, for selected patients, it can be an acceptable alternative to inpatient care, but recruited volumes were low and the case for substituting hospital beds wholesale is not established (Edgar et al., 2024). Admission avoidance works for some people, some of the time; it is not a lever that can simply be pulled at scale.

Perhaps the most striking finding for home care providers concerns something far less technological: continuity of care. A systematic review of continuity and mortality found that 18 of 22 higher-quality studies, across nine countries, reported statistically significant reductions in mortality where patients experienced greater continuity of care (Pereira Gray et al., 2018). The authors are appropriately cautious (the evidence is observational, so it demonstrates association rather than proven causation), but the consistency is remarkable, and it suggests that simply having the same familiar people involved over time is itself a form of prevention. In domiciliary care, continuity is operational: it is a function of how visits are allocated and who turns up. That makes it one of the few preventative levers an agency directly controls.

Routine care data as an early-warning system

The distinctive asset of home care is that carers see people repeatedly, in their own environment, in a way no clinic can. A carer notices that Mrs Okafor is eating less, that the flat is colder than usual, that she seems more confused than last week, that the same bruise has appeared twice. Historically that knowledge evaporated, trapped in one carer's head or an unread paper note. Structured, routinely-collected care data changes the arithmetic: if observations are captured consistently, small signals can be aggregated into a picture of change that no single visit would reveal.

This is where population health management theory meets the visit-level reality of domiciliary care. NHS England's model depends on data to segment and target (NHS England, 2025); home care is, potentially, one of the richest and least-exploited data sources feeding that model, because it observes the frail population precisely where deterioration begins: at home, between appointments. The prize is an early-warning system built not from expensive new sensors but from the observations carers already make on every visit. The condition, as with everything above, is that the data is captured well enough to be trusted and that someone is positioned to act on it.

How homecareOS is built for a preventative model

A note on where our own platform fits. homecareOS is designed around exactly this idea: that routine visit data, captured well, can become an early-warning signal. Its digital care records turn each visit into structured data (observations, body-map annotations for skin and wound changes, and outcome-focused goals) rather than free text that is written once and never read again. Because care records are immutable and time-stamped, patterns become visible over time instead of vanishing between carers.

The continuity evidence (Pereira Gray et al., 2018) directly informs how the platform allocates work. The cover finder deliberately weights continuity of care (familiarity between carer and service user) as a major factor when a visit needs covering, alongside distance and hours, so that finding cover quickly does not quietly erode the very continuity that appears to protect people. We are honest that this evidence is associational, not proven cause, and we do not overclaim from it.

On the analytics side, homecareOS turns those routine observations into reports and dashboards (including absence-pattern indicators such as the Bradford Factor and a composite retention-risk signal) that surface change early enough to act. We are equally clear about what this is not: it is not a validated clinical predictor of an individual's falls or deterioration, and, consistent with NICE guidance (NICE, 2025), we do not present it as one. The research base that shapes these choices is set out in full on our evidence page. And because this essay is about the care model rather than a product's AI checklist, readers who want the feature-by-feature reality of what is genuinely shipped versus roadmap should see our separate review of AI in home care software for 2026, alongside our guide to comparing home care software in the UK.

The honest limits of prediction

A responsible account of predictive care has to dwell on its failure modes, because the enthusiasm frequently outruns the evidence. Three limits recur throughout the studies above.

False positives and induced demand. A risk score that is acted on indiscriminately can do harm, generating investigation, anxiety and admission without benefit, precisely the pattern PRISMATIC observed (Snooks et al., 2018). Every prediction carries a base rate; in a population where most people will not fall or deteriorate this week, even an accurate-sounding tool produces many false alarms, each with a human and financial cost.

Data quality. Predictive models are only as good as the observations feeding them. In domiciliary care, data is captured by busy people between visits, sometimes offline, sometimes in a hurry. Inconsistent recording produces inconsistent signals, and a model trained on incomplete or biased inputs will mislead confidently. Prediction is downstream of data discipline, not a substitute for it.

Ethics and governance. Reusing personal data to predict and profile people raises real questions of transparency, bias and accountability, as the Nuffield Council on Bioethics has set out for AI and data-driven tools in health and care (Nuffield Council on Bioethics, 2018). Who sees a risk flag, how it is explained to the person it concerns, whether it entrenches existing inequities, and how the underlying data is governed are not afterthoughts: they are preconditions for doing this well. We treat these questions at greater length in our companion essay on the ethics of care technology.

Where this leaves home care

The direction of travel is not in serious doubt: an ageing population living longer in poor health (ONS, 2024), a workforce that cannot expand fast enough to meet demand (Skills for Care, 2024), and a policy consensus around proactive, data-driven care (NHS England, 2025) all point the same way. What the evidence tempers is the fantasy that prediction alone will get us there. Risk-stratification tools without a plan can make things worse (Snooks et al., 2018); falls predictors are not recommended for individuals (NICE, 2025); monitoring pays off only inside a responsive service (Steventon et al., 2012); and admission-avoidance schemes work for some, not all (Edgar et al., 2024).

The most defensible reading is that the biggest preventative gains in home care will come less from exotic prediction and more from doing ordinary things consistently: capturing observations well, maintaining continuity of care (Pereira Gray et al., 2018), and building the routine data that lets an agency notice change early. That is an unglamorous conclusion, and it is probably the right one. For the wider context, see the pillar of this series, the future of UK home care.

References

1. Office for National Statistics (2024) Healthy life expectancy in England and Wales: between 2011 to 2013 and 2021 to 2023. London: ONS. Released 12 December 2024. View source
2. Skills for Care (2024) The state of the adult social care sector and workforce in England 2024. Leeds: Skills for Care. Published 10 October 2024. View source
3. Clegg, A., Bates, C., Young, J., Ryan, R., Nichols, L., Teale, E.A., Mohammed, M.A., Parry, J. and Marshall, T. (2016) ‘Development and validation of an electronic frailty index using routine primary care electronic health record data’, Age and Ageing, 45(3), pp. 353-360. View source
4. NHS England (2023, updated 2025) Population health management. NHS England long read. First published 17 February 2023; last updated 4 April 2025. View source
5. Snooks, H., Bailey-Jones, K., Burge-Jones, D., et al. (2018) ‘Predictive risk stratification model: a randomised stepped-wedge trial in primary care (PRISMATIC)’, Health Services and Delivery Research, 6(1). Southampton: NIHR Journals Library. View source
6. National Institute for Health and Care Excellence (2025) Falls: assessment and prevention in older people and in people 50 and over at higher risk. NICE guideline NG249. Published April 2025. View source
7. Steventon, A., Bardsley, M., Billings, J., Dixon, J., Doll, H., Hirani, S., et al. (2012) ‘Effect of telehealth on use of secondary care and mortality: findings from the Whole System Demonstrator cluster randomised trial’, BMJ, 344:e3874. View source
8. Hamblin, K. and Whitfield, G. (2025) On hold: the Analogue-to-Digital Switchover and telecare. Centre for Care Working Paper 5. Sheffield: Centre for Care. May 2025. View source
9. Edgar, K., Iliffe, S., Doll, H.A., Clarke, M.J., Goncalves-Bradley, D.C., Wong, E. and Shepperd, S. (2024) ‘Admission avoidance hospital at home’, Cochrane Database of Systematic Reviews, Issue 3, Art. No. CD007491. View source
10. Pereira Gray, D.J., Sidaway-Lee, K., White, E., Thorne, A. and Evans, P.H. (2018) ‘Continuity of care with doctors: a matter of life and death? A systematic review of continuity of care and mortality’, BMJ Open, 8(6):e021161. View source
11. Nuffield Council on Bioethics (2018) Artificial intelligence (AI) in healthcare and research. Policy briefing, 15 May 2018. London: Nuffield Council on Bioethics. View source