For twelve years I studied what changes in a brain when you train it. Cognitive aging, neuroplasticity, music as intervention, structural connectivity by diffusion MRI. The work was good, the methods were careful, and the finding that kept returning was the one I could not explain inside the field I was working in.
People with similar pathology had different trajectories. People with comparable training protocols showed wildly different gains. The variance was not in the brain. The variance was in the conditions the brain was living inside, and the field had no instruments for measuring those conditions.
This is a field note on what I learned, and on why I built SEAM as the diagnostic apparatus the field is missing.
What The Models Could Not Hold
Neuroplasticity is real. The brain changes when you give it the right input over enough time. The molecular, cellular, and systems-level mechanisms have been mapped across two decades of careful work (Bavelier, Levi, Li, Dan, & Hensch, 2010; Lövdén, Bäckman, Lindenberger, Schaefer, & Schmiedek, 2010). Cognitive aging shows similar plasticity properties. The aging brain compensates, reorganizes, and continues to learn well into the eighth and ninth decades when conditions support it (Park & Reuter-Lorenz, 2009).
What kept showing up in my data and in the broader literature was a problem the plasticity framework alone could not solve. Two participants with comparable baseline cognition, comparable structural connectivity by DTI, comparable training protocols, would respond completely differently. One would consolidate gains and carry them forward. The other would show a transient bump and then drift back to baseline within months. The training was the same. The brain measures were the same. The outcomes diverged.
The variance had to live somewhere. The field's instinct was to look for it inside the head. More refined connectomics. Better diffusion models. Higher-resolution cortical mapping. These are all good methods and they did not close the gap.
Plasticity is relationally gated. The same intervention produces durable change in one nervous system and fragile change in another, and a large fraction of that difference lives upstream of any cognitive variable we routinely measure. The relational scaffold a person is held inside is not a soft covariate. It is the substrate that determines whether a clinical gain holds.
I would sit with two scans, two cognitive profiles, two training records, and the question of why one person consolidated and the other did not. The answer was rarely in the file. It was almost always in the parts of the person's life the file did not contain, the parts the recruitment form had not thought to ask about.
The Variables That Were Never in the Model
Years of education was always in the model. Socioeconomic status was always in the model. Lifestyle variables, sometimes meditation status, sometimes physical activity, sometimes diet quality. What was never in the model was the question of why a person was taking anxiety medication. Whether they had been through a recent divorce. Whether the home they were returning to after the scan was a place they could rest. Who they were connected to. Whether those connections supported them or drained them. Whether they belonged to anything at all.
The evidence that these variables matter is not new. Berkman and Syme's nine-year follow-up in Alameda County showed that adults with fewer social ties had mortality rates two to three times higher than adults with more ties, after adjusting for age, baseline health, socioeconomic status, and behavioural variables (Berkman & Syme, 1979). The finding has been replicated across decades and across cohorts. Holt-Lunstad's meta-analysis of 148 studies found that the effect size of social relationships on mortality was comparable to, and in some cases larger than, established risk factors like smoking and obesity (Holt-Lunstad, Smith, & Layton, 2010).
Social rejection and exclusion are processed by partially overlapping neural systems with physical pain. Eisenberger's work demonstrates that social disconnection engages affective pain and salience-related regions, particularly the dorsal anterior cingulate cortex and anterior insula, in ways that share substrate with the affective dimension of physical pain (Eisenberger, 2012). Later multivariate analyses qualified this picture: despite the regional overlap, physical pain and social rejection can be distinguished at finer levels of neural representation, and the systems are best described as shared but not identical (Woo et al., 2014). Chronic social stress contributes to allostatic load through these and related pathways (McEwen, 2007), even where the two responses diverge in their fine-grained representations.
Knowing all of this, neuroscience studies still routinely exclude relational variables from the models. We adjust for education. We adjust for income. We do not ask the participant whether they have someone to call when they are afraid. We do not measure the density of their social network or the integrity of their primary attachments. We design studies that treat the participant as if their brain arrived in the scanner detached from the system that produced it and the system it returns to.
This is a severe limitation in current neuroscience. We measure the brain in the scanner. We assess the person against thin demographic categories. The science we are doing is missing a load-bearing variable, and we have built that omission into the architecture of how studies get designed and published.
My own doctoral years were spent inside the COVID-19 pandemic. My cohort lived through forced separation, prolonged distance from primary attachments, and the collapse of routine relational structures that had been load-bearing for them. I lived through it with them. None of this was in the models. The variance the field was struggling to explain was being produced, in real time, by a relational shock the size of a generation. The shock was not measured. The data were collected and analyzed as if the conditions of everyday life had not just been disrupted at a scale not seen in living memory.
I sat with that gap for a long time before I named what to do with it.
From Brain to Bond to Network
The pivot, when it came, was through health systems thinking. The pattern showed up once I started looking at how care actually moves between providers and across settings. Communication failure was the thing producing the poor quality of care that patients experienced. The literature on integrated care described what should happen (Wagner, 1998; Bodenheimer, Wagner, & Grumbach, 2002; Katon et al., 2010). What actually happened looked different.
The integration was structural in name and fragmented in practice. The dyads worked. A patient and their primary care physician could have a good relationship. A patient and their therapist could have a good relationship. The breakdown was in the architecture connecting those dyads. Communication failed between them. Data did not move between them. Handoffs dropped. The wraparound that looked like one network on the org chart turned out to be a loose collection of nodes with very few measured edges between them, and the patient felt every one of those missing edges as a degradation in the quality of care they received.
That observation moved me from brain to bond and from bond to network. The unit of analysis I needed was not the patient and not the dyad. It was the structure connecting both, and the structure has properties of its own that are not reducible to the properties of the nodes inside it (Bullmore & Sporns, 2009).
Care Networks That Cannot Hold
Collaborative care fails at the joints. Coordination fails. Consensus building fails. Quality standards do not hold across settings. Data does not move between systems that have legitimate reasons not to share it and no shared apparatus for doing it safely. A patient passes from emergency to inpatient to outpatient to community, and the relational thread that should carry context across those handoffs is asked to live in chart notes that no one has time to read.
I love applied distributed systems work because it gives language to what is actually happening here. Distributed systems theory has spent fifty years developing precise terminology for a class of problems that collaborative care is also a member of (Lamport, 1978). Many care networks are not fault tolerant. They have no redundancy at their critical joints. When one node goes down, the work that node was doing does not get picked up by another node, and the patient experiences this as being dropped.
The networks are not resilient. Holling's foundational work defined resilience as the capacity of a system to absorb disturbance and reorganize while undergoing change so as to still retain essentially the same function, structure, and identity (Holling, 1973). Care networks designed without resilience properties have no slack. They cannot absorb load. A single staff change, a single funding cut, a single missed referral can produce cascading failure for the people relying on them.
Single points of failure are not anomalies in these systems. They are structural features the field has stopped seeing because the framing of care as a service delivery problem hides the network architecture underneath. Ostrom's work on polycentric governance offered an alternative architectural model for managing common-pool resources, one with multiple overlapping centers of decision-making and built-in mechanisms for adaptation (Ostrom, 1990). Most care networks were not designed with these properties because the language for designing them was not in the room when the protocols were written.
What gets discussed at care team meetings is the patient's diagnosis, the patient's symptoms, the patient's adherence. What rarely gets discussed is whether the network of providers in the room can actually function as a network when no one is in the room together. The protocol assumes integration that the structure has not been designed to produce.
Fragmentation in collaborative care is a structural problem dressed up as a coordination problem. Treating it as coordination produces meetings, protocols, and handoff forms. Treating it as structure produces something different. It produces measurement. It produces diagnostics. It produces a way to ask whether a care network has the integrity to hold a person through what they are about to go through.
The Public Health Recognition Has Arrived
The evidence base has been moving. Public health bodies have begun to say what the literature has shown for decades.
In May 2023, the US Surgeon General released an advisory titled Our Epidemic of Loneliness and Isolation (Office of the Surgeon General, 2023). The advisory framed lack of social connection as a public health crisis comparable to tobacco use and obesity. About half of US adults reported experiencing loneliness. The mortality risk associated with social disconnection was estimated as equivalent to smoking up to fifteen cigarettes a day. The advisory drew on earlier analysis estimating that social isolation among older adults accounts for approximately $6.7 billion in additional Medicare spending annually (Flowers et al., 2017).
In June 2025, the World Health Organization Commission on Social Connection released its flagship report, From Loneliness to Social Connection: Charting a Path to Healthier Societies (WHO Commission on Social Connection, 2025). The findings are stark. One in six people worldwide is affected by loneliness. Loneliness contributes to more than 871,000 deaths annually, roughly one hundred deaths every hour. The Commission proposes that social connection be recognized as a third pillar of health, alongside mental and physical health. This is a conceptual proposal from the Commission and its supporters, not yet an established WHO classification. The World Health Assembly adopted its first-ever resolution on social connection in May 2025, urging member states to develop evidence-based policies to address it.
Holt-Lunstad's 2024 review in World Psychiatry consolidated the evidence (Holt-Lunstad, 2024). Social connection factors are independent predictors of mental and physical health. Some effect sizes are comparable to or exceed established risk factors like smoking, obesity, and physical inactivity.
The recognition exists. The clinical apparatus to act on it does not.
A 2025 survey of 681 healthcare providers asked them to rank the importance of various health factors for mortality and chronic illness (Holt-Lunstad et al., 2025). When their rankings were compared against the actual effect sizes in the published literature, social connection ranked low in perceived importance relative to its actual contribution. Providers identified lack of time, lack of resources, lack of training, and lack of confidence as the significant barriers to addressing social connection in clinical settings. The authors concluded that integration of social connection into healthcare requires educational programs, institutional policies, and structural changes that do not currently exist.
This is the gap the field is now standing in. The evidence is in. The advisories have been published. The infrastructure to act on any of it has not been built.
Relational Well-Being Is Public Health
The advisories have arrived. The frame they are working within is still incomplete. They name loneliness as a risk factor. They have not yet named cohesion as a structural property to measure. Cohesion across scales is the load-bearing structure that determines whether any clinical intervention will hold once the patient leaves the building.
Marmot's work on the social determinants of health established two decades ago that the conditions in which people are born, grow, live, work, and age account for more variance in health outcomes than the healthcare system itself (Marmot, 2005). Berkman and colleagues have extended this through social epidemiology, mapping how relational structures at multiple levels mediate the conditions that produce health (Berkman, Kawachi, & Glymour, 2014). The evidence base is large. The integration of this evidence into how care is designed and delivered is small.
A health plan that treats the body, treats the mind, addresses lifestyle, and ignores the relational integrity of the systems the person lives inside is an incomplete plan. The biomedical model has been working very hard for fifty years to add layers to itself. Biopsychosocial. Behavioural. Lifestyle. Each layer expanded the model and each layer stopped short of measuring the thing that holds all of them together.
The relational layer is not optional. It is not adjunctive. It is not what you do if you have time and money left over after the clinical work. It is the infrastructure the clinical work sits on. Without it, the clinical work fails by attrition. People stop taking medication because no one notices they stopped. People stop showing up to therapy because the bus did not come and no one called. People relapse because the family system they returned to was the system that made them sick. None of this is mysterious. We have known it for a long time. What we have not done is build the instruments that let us see it as a measurable property of a care system. We have left it as private suffering happening offstage from the real medicine.
Relational well-being is public health. I want that line carried out of this issue and into every conversation that happens next.
The Architecture That Holds the Question
SEAM did not emerge from any one of the disciplines I trained in. It emerged from sitting between them long enough to see what each one missed and what each one could contribute when held in relation to the others.
The intellectual architecture of HILI is organized around this orientation. Systems thinking sits at the top as the integrative lens. Below it, nine substrate disciplines produce the theoretical claims: cognitive and affective neuroscience, network science, information theory, systems sociology, family systems theory, dynamical systems, public health, and distributed systems. Between substrate and concept sit six methodological approaches that carry the same mathematical apparatus across very different empirical objects: computational modeling, computational social science, social network analysis, graph theory, signal processing, and topological data analysis. The same instruments that map white matter tracts in a brain map relational ties in a care network. The substrate shifts. The methods do not.
Below the methods sit the operational concepts the field uses in papers and clinical conversations. SEAM sits below them as the integrative framework that holds them in relation to each other. Below SEAM sit the application domains where the framework gets used.
The architecture is not decorative. Each level was added because the work required it. Neuroscience needed sociology to explain the variance it kept seeing. Sociology needed network science to formalize the structure it was describing. Network science needed distributed systems and topological tools to handle multi-scale relational data. Public health needed an apparatus that crossed clinical settings to act on the evidence already in front of it. The diagram is what HILI looks like when you ask each discipline what it needs from the others to answer the question that matters.
Why SEAM Had to Exist
SEAM is the instrument I built because I could not find one that already existed. Five dimensions, the same five at every scale that matters: Empowerment, Growth, Freedom, Support, Commitment. A dyad has these properties. A family has these properties. A team has these properties. A care network has these properties. A community has these properties.
The scale invariance is the structural argument. The same five dimensions diagnostically work whether you are looking at a marriage or a multi-agency wraparound program. The mathematics of network analysis says this. The substrate shifts and the topology repeats.
SEAM exists because I needed a way to ask whether a system, at any scale, had the integrity to hold a person through what they were about to go through. The biomedical apparatus could not answer that question. The collaborative care literature gestured at it without measuring it. The neuroscience I came from did not ask it.
The variance the brain research could not explain, the patients the care networks could not hold, the variable the public health frame keeps stopping short of measuring: these are not separate problems. They are one problem viewed at different scales, and the scale-invariant property they all point at is relational integrity.
I built SEAM because the apparatus did not exist and the people who needed it were being failed by its absence. The instrument is built. What remains is the work of changing what this field is willing to ask.
References and Theoretical Grounding
Bavelier, D., Levi, D. M., Li, R. W., Dan, Y., & Hensch, T. K. (2010). Removing brakes on adult brain plasticity: From molecular to behavioral interventions. Journal of Neuroscience, 30(45), 14964–14971.
Berkman, L. F., Kawachi, I., & Glymour, M. M. (Eds.). (2014). Social Epidemiology (2nd ed.). Oxford University Press.
Berkman, L. F., & Syme, S. L. (1979). Social networks, host resistance, and mortality: A nine-year follow-up study of Alameda County residents. American Journal of Epidemiology, 109(2), 186–204.
Bodenheimer, T., Wagner, E. H., & Grumbach, K. (2002). Improving primary care for patients with chronic illness. JAMA, 288(14), 1775–1779.
Bullmore, E., & Sporns, O. (2009). Complex brain networks: Graph theoretical analysis of structural and functional systems. Nature Reviews Neuroscience, 10(3), 186–198.
Eisenberger, N. I. (2012). The pain of social disconnection: Examining the shared neural underpinnings of physical and social pain. Nature Reviews Neuroscience, 13(6), 421–434.
Flowers, L., Houser, A., Noel-Miller, C., Shaw, J., Bhattacharya, J., Schoemaker, L., & Farid, M. (2017). Medicare spends more on socially isolated older adults. AARP Public Policy Institute.
Holling, C. S. (1973). Resilience and stability of ecological systems. Annual Review of Ecology and Systematics, 4, 1–23.
Holt-Lunstad, J. (2024). Social connection as a critical factor for mental and physical health: Evidence, trends, challenges, and future implications. World Psychiatry, 23(3), 312–332.
Holt-Lunstad, J., Proctor, A. S., Perissinotto, C. M., Cheng, A., Cudjoe, T. K. M., Kotwal, A. A., & Morley, T. (2025). Healthcare providers’ perceived importance and barriers to addressing social connection in medical settings. Annals of the New York Academy of Sciences.
Holt-Lunstad, J., Smith, T. B., & Layton, J. B. (2010). Social relationships and mortality risk: A meta-analytic review. PLoS Medicine, 7(7), e1000316.
Katon, W. J., Lin, E. H. B., Von Korff, M., Ciechanowski, P., Ludman, E. J., Young, B., Peterson, D., Rutter, C. M., McGregor, M., & McCulloch, D. (2010). Collaborative care for patients with depression and chronic illnesses. New England Journal of Medicine, 363(27), 2611–2620.
Lamport, L. (1978). Time, clocks, and the ordering of events in a distributed system. Communications of the ACM, 21(7), 558–565.
Lövdén, M., Bäckman, L., Lindenberger, U., Schaefer, S., & Schmiedek, F. (2010). A theoretical framework for the study of adult cognitive plasticity. Psychological Bulletin, 136(4), 659–676.
Marmot, M. (2005). Social determinants of health inequalities. The Lancet, 365(9464), 1099–1104.
McEwen, B. S. (2007). Physiology and neurobiology of stress and adaptation: Central role of the brain. Physiological Reviews, 87(3), 873–904.
Office of the Surgeon General. (2023). Our Epidemic of Loneliness and Isolation: The U.S. Surgeon General’s Advisory on the Healing Effects of Social Connection and Community. US Department of Health and Human Services.
Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press.
Park, D. C., & Reuter-Lorenz, P. (2009). The adaptive brain: Aging and neurocognitive scaffolding. Annual Review of Psychology, 60, 173–196.
Wagner, E. H. (1998). Chronic disease management: What will it take to improve care for chronic illness? Effective Clinical Practice, 1(1), 2–4.
Woo, C.-W., Koban, L., Kross, E., Lindquist, M. A., Banich, M. T., Ruzic, L., Andrews-Hanna, J. R., & Wager, T. D. (2014). Separate neural representations for physical pain and social rejection. Nature Communications, 5, 5380.
World Health Organization Commission on Social Connection. (2025). From loneliness to social connection: Charting a path to healthier societies. World Health Organization.