When did someone last ask you who you would call at two in the morning?
It’s not a question on any standard intake form, yet according to Florence Lewis, PhD, LMFT, President and Founder of Upside Health Research Network, it may carry more clinical weight than half the questions that are. Lewis, a Medical Family Therapist, a licensed Marriage and Family Therapist trained to work in medical settings, who trains healthcare teams on bringing AI into care without losing the relationship at its centre, argues that relational health has become a missing metric in modern medicine, not because clinicians don’t care, but because nobody ever trained them to ask, and the visit is short.
Healthcare already measures a great deal well: blood pressure, weight, A1c, and screeners like the PHQ-9. What’s missing sits beside them: a structured way of capturing whether a patient has anyone to call when things go wrong.
Loneliness is a risk factor, not a mood
Loneliness belongs in the same category as high blood pressure or smoking, Lewis argues, not filed away as a passing feeling. It’s also distinct from a related concept: social isolation is objective, how much contact a person actually has, while loneliness is subjective, the gap between the connection someone wants and what they have. Someone can be isolated and content, or surrounded by people and lonely. Screening for one will miss the other.
The evidence is substantial. A 2015 meta-analysis pooling 70 studies found social isolation, loneliness, and living alone were each associated with a 26–32% higher risk of dying during follow-up, even after adjusting for standard health confounders. In 2024, the Lancet Commission on dementia named social isolation among fourteen modifiable risk factors for the condition, with modifiable meaning clinicians can act on it. In the US, Healthy People 2030 already lists social and community context as a recognised social determinant of health. Lewis is upfront that most of the mortality evidence is observational, showing pattern rather than proof of cause. Still, she says, a risk factor this consistent is worth one line on a form.
The question that beats every intake form
Translating this into practice is smaller than most clinicians expect, Lewis says. Cohen and Wills drew the relevant distinction back in 1985: structural social support is the shape of a network, while functional support is whether it actually delivers when it matters. In practice, that’s the difference between asking who a patient lives with and asking who they’d call at two in the morning, and the second question usually carries the real information. When the answer is nobody, the response isn’t to fix it in that visit. It goes in the care plan.
There’s already precedent for treating a support network as clinically consequential: federal rules for ambulatory surgery centres (42 CFR § 416.52) require patients to be discharged in the company of a responsible adult. It’s simply never been logged as a health indicator more broadly.
4 Things worth documenting instead of “Do you have a good support system?”
Almost every patient says yes to that question. Lewis suggests four more useful things to document instead:
- Perceived availability of support: who a patient actually believes would show up, as distinct from a listed emergency contact.
- Quality over network size: a large, high-conflict network offers no protection. A 2022 pooled analysis by Mann and colleagues, published in Social Psychiatry and Psychiatric Epidemiology, found frequent loneliness carried more than double the adjusted odds of new-onset depression, with a caveat the researchers themselves flagged: notable variation across studies and an older-adult skew.
- Reciprocity: many patients carry real caregiver burden with none of their own.
- A follow-up at six months: one data point is a snapshot; two form a trend.
The DSM-5-TR Cultural Formulation Interview already includes a relevant section, asking what makes a problem better or worse, naming family and friends directly.
Why support looks different across cultures
Screening only works if it recognises support looks different across communities. A form that only asks about a spouse or partner misses chosen family, a term from anthropologist Kath Weston’s 1991 work on LGBTQ+ kinship, now treated clinically as real support rather than sentiment.
Lewis separates cultural context from cause carefully: the elevated risk some patients carry lives in what they’re exposed to, not in who they are. For LGBTQ+ patients, that’s minority stress, a term Ilan Meyer coined in 2003 for the ongoing weight of stigma and expected rejection. For patients of colour, Shelly Harrell’s related but distinct concept is racism-related stress, the parallel weight of navigating racism day to day. Arline Geronimus’s research on “weathering”, tracking Black and white respondents specifically, found long-term stress exposure shows up as earlier health decline, independent of income.
The picture gets more specific still. A 2024 study of national data on older adults shows the Black-white loneliness gap largely disappears once income and education are accounted for, and Latino older adults report the least loneliness of any group studied. A Kaiser Family Foundation survey found people facing discrimination several times a year reported loneliness at 26%, against 7% among those who rarely did, though the survey design shows association, not that discrimination causes loneliness. “This was never about race,” Lewis says. “It’s about exposure.”
Lewis is careful not to frame any of this only as deficit. She points to cultural humility, a framework proposed by Tervalon and Murray-García in 1998, as an ongoing willingness to question your own assumptions rather than a finished credential, the better standard for every form and every tool: who does this leave out? And she’s just as insistent on naming what these communities have built in response. Researcher Karen Fredriksen-Goldsen titled her landmark study of LGBTQ+ older adults Disparities and Resilience, choosing both words on purpose. These are communities that built support systems from scratch when the standard ones failed them, and that strength belongs in the clinical picture too.
One case has stayed with her: a pregnant patient, a woman of colour, carrying a significant history of childhood trauma. Every risk factor lined up in Lewis’s head at once, she says, until the patient described, unprompted, everything she’d already done to protect her own mental health. Near the end, she told Lewis: “Social support is like medicine for me.” Reading the mortality research later, Lewis realised she hadn’t been speaking figuratively.
What AI can and can’t do here
Lewis’s current work explores how AI can support tracking relational health without replacing the clinical relationship itself. Relational health isn’t static. A divorce, a death, or a caregiving load can land overnight, and no clinician can hold that across a full patient panel. Software can, and longitudinal tracking frees up time to spend looking a patient in the eye rather than at a screen.
But her research also studies behavioural health AI for bias, pointing to a concept from psychometrics called measurement invariance, described by Putnick and Bornstein in 2016: whether a tool measures the same thing, the same way, across different groups. Most relational health tools haven’t been tested to that standard, which makes cross-group score comparisons a guess dressed up as data. Her test for any tool: does it pull the human moment forward, or push it aside?
This is the foundation of the continuing education Upside now offers counsellors and mental health clinicians, covering AI in mental health care, relational health as a vital sign, and connection across the lifespan.
Upside Health Research Network has been approved by NBCC as an Approved Continuing Education Provider, ACEP No. 8132. Programs that do not qualify for NBCC credit are clearly identified. Upside Health Research Network is solely responsible for all aspects of the programs.
What would change in the next twenty visits, Lewis asks, if there were one better question in the room?
Florence Lewis, PhD, LMFT, has spent her career on a simple premise: the relationships in a patient’s life are as much a clinical indicator as their blood pressure. As president and founder of Upside Health Research Network, a non-profit research institute, she now works with healthcare systems worldwide to bring that thinking and AI into care, without losing the relationship at its centre.
