Can AI profiles tell nurses what support a patient needs? Most evidence stops before real-world care
A bilingual scoping review included 61 studies, but many evaluated algorithms, feasibility or system development rather than whether a valid profile led to safer, fairer or more effective nursing support.
The short version
- What happened
- The review searched four Chinese and four English-language databases through 18 May 2026 and included 61 studies from 1,172 records.
- Who it affects
- Nurses may be asked to act on labels assembled from sensitive clinical and behavioural data. Patients need to know what the profile changes, who can correct it and whether missing digital data can reduce access to support.
- What remains unproven
- Scoping review
Keep in mind
Scoping review
61 heterogeneous studies
Retrospective protocol registration
No formal risk-of-bias appraisal
No effectiveness meta-analysis
Grey literature excluded
Few long-term patient-centred outcomes
Inconsistent equity assessment
Source analysis · 1 linked source. Primary analysed source date: 29 October 2026.
Read the full report · 4 min The evidence, methods, implications and limitations
The field contains many kinds of 'profile'
A new scoping review maps a broad family of systems that turn patient-related data into profiles intended to support nursing or personalised health care. The researchers searched PubMed, CINAHL, Web of Science, Embase and four Chinese databases through 18 May 2026. From 1,172 records, they assessed 106 full texts and included 61 studies. Two reviewers independently screened and charted the evidence using a predefined form.
The label covered patient profiles and personas, digital phenotypes, clinical subtypes, segmentation, digital twins and large-language-model personalisation. Inputs ranged from health records and questionnaires to mobile, wearable, sensor and conversational data. Outputs were used or proposed for education, self-management, rehabilitation, chronic-disease support, monitoring, follow-up and shared decisions. Combining these forms shows breadth, but it also means the studies did not test one common intervention.[1]
A profile is useful only if it leads to a defensible action
The authors classified 31 studies as directly relevant to nursing or nurse-relevant care, 15 as transferable health support and 15 as methodological evidence with plausible future relevance. Most evaluated algorithm performance, feasibility, usability or system development. Fewer examined whether a profile was valid, reproducible and actionable; whether nurses could apply its recommendation consistently; or whether patients experienced better outcomes over meaningful follow-up.
That gap matters in everyday care. A cluster labelled 'low adherence' or 'high need' can change who receives education, monitoring or a follow-up call. If the profile is unstable, shaped by missing digital data or poorly matched to a care pathway, it can reinforce inequity rather than personalise support. Procurement should therefore require an explicit rule connecting each profile to an action, evidence that staff can understand and override it, and testing across language, disability, age, health-literacy and access groups.[1]
The distinction that matters
From patient data to care—where evidence thins
A profile becomes clinically meaningful only when each step is defined and tested.
Input
Records, surveys, sensors or conversations
Coverage and missingness can differ across patients
Profile
Segment, phenotype, persona or digital twin
Validity and transferability were often uncertain
Action
Education, monitoring, follow-up or support
Long-term nursing and patient outcomes were less often tested
The review maps the field; it does not prove effectiveness
The review did not conduct a formal risk-of-bias appraisal, meta-analysis or intervention-effectiveness assessment. Its protocol was registered after study selection and data charting had begun, so the registration is retrospective documentation rather than a prospective constraint. The literature was heterogeneous, terminology was inconsistent, and grey literature and preprints were excluded. Those choices make the map inspectable but prevent a claim that 61 studies amount to mature clinical evidence.
The work was supported by Chinese university and disease-surveillance research grants; the authors said the funders had no role and declared no conflicts. Confidence would rise with independently validated profiles, pre-specified profile-to-action rules and prospective comparisons against ordinary nursing practice that measure workload, missed needs, inappropriate escalation, patient experience, safety and unequal effects.[1]