Research Methods in Primary Care: Designing for the Setting You Have

The evidence that governs primary care was produced in the one setting where almost no primary care happens.

In 1961, Kerr White, T. Franklin Williams, and Bernard Greenberg published "The Ecology of Medical Care" in the New England Journal of Medicine. They asked where people with medical problems actually are in a given month. Of a thousand adults, about 750 experienced an episode of illness or injury. About 250 consulted a physician. Nine were admitted to a hospital. One was referred to a university medical center.

Forty years later Larry Green, George Fryer, Barbara Yawn, David Lanier, and Susan Dovey repeated the count in the same journal. "The Ecology of Medical Care Revisited" found the proportions essentially unchanged. Of a thousand people in a month, roughly eight hundred report symptoms, a little over two hundred visit a physician's office, thirteen visit an emergency department, eight are hospitalized, and fewer than one is hospitalized in an academic medical center.

Fewer than one in a thousand. That is the setting that generates the trials, writes the guidelines, trains the investigators, and holds the funding.

This is usually described as a dissemination problem. Good evidence exists, the familiar figure of a seventeen-year lag from publication to practice applies, and the fix is implementation. That framing is comfortable because it puts the failure downstream of the science. It is also wrong. The problem starts at design. A study built inside an academic medical center encodes that center's conditions in its eligibility criteria, its follow-up schedule, its outcome measurement, and its staffing assumptions. When the study closes, its answer belongs to those conditions. Carrying it to a tribal clinic in northeast Nebraska or a federally qualified health center in south Omaha is not dissemination. It is extrapolation, and nobody labels it that way.

Issue 010 of this publication argued that limited-resource medicine is the actual practice of medicine for tens of millions of Americans, in places it named directly: Pine Ridge, the Mississippi Delta, Appalachia, the FQHCs, the tribal health systems. The research consequence follows from that argument. Those settings are largely absent from the evidence base that governs them, and the clinicians working in them are told, whenever the evidence fits badly, that they are practicing outside it.

They are not practicing outside the evidence. The evidence was never generated for them.

The methods that fit those conditions already exist. Several were built by people working under exactly the constraints the academic trial cannot tolerate.

Why the Trial Does Not Travel

Daniel Schwartz and Joseph Lellouch named the distinction in 1967, in "Explanatory and pragmatic attitudes in therapeutical trials" in the Journal of Chronic Diseases. An explanatory trial asks whether an intervention can work under conditions arranged to give it the best chance. A pragmatic trial asks whether it does work under the conditions in which it will be used. Both questions are legitimate. They are not the same question, and an answer to the first is not an answer to the second.

Most trials that produce primary care guidelines lean explanatory. The consequences are specific and they compound.

Eligibility comes first. Harriette Van Spall and colleagues sampled randomized trials published in high-impact general medical journals and reported in JAMA in 2007 that exclusions for comorbidity, concurrent medication, and age were routine and frequently unexplained. The patient with diabetes, hypertension, chronic kidney disease, depression, and an untreated substance use disorder is not an unusual patient in a community health center exam room. She is close to the median patient in that room. She would not have been enrolled in the study that produced her treatment plan.

Then the guidelines stack. Cynthia Boyd and colleagues, writing in JAMA in 2005, took a hypothetical older woman with five common chronic conditions and applied the relevant disease-specific guidelines to her at once. The result was roughly a dozen medications taken at five separate times of day, at a monthly drug cost in the hundreds of dollars. Each guideline was defensible on its own evidence. The sum was not a treatment plan.

Then the logistics. A protocol requiring four follow-up visits in twelve weeks assumes transportation, a stable phone number, an employer who tolerates absence, and childcare. A trial that measures its primary outcome with an assay the clinic does not run assumes a lab. A trial that requires a study coordinator assumes a study coordinator. These are not incidental features of a design. They are its load-bearing structure, and they select the sites where the design can run at all.

Kirsty Loudon and colleagues built the PRECIS-2 tool, published in the BMJ in 2015, to make those decisions visible before enrollment rather than after publication. It scores a trial on a continuum from explanatory to pragmatic across nine domains: eligibility, recruitment, setting, organization, flexibility of delivery, flexibility of adherence, follow-up, primary outcome, and primary analysis. Its value is not as a critique of existing trials. It is a design instrument. It forces the investigator to state which question is being asked and for which setting the answer will hold.

Methods Built for Constraint

Practice-based research networks solve a problem the academic center does not have and cannot fix. A primary care practice knows its panel. It has a denominator. Referral-based academic medicine sees a filtered population, which is why prevalence estimates drawn from tertiary centers are so reliably wrong for the community. At the end of the 1970s American family medicine organized the Ambulatory Sentinel Practice Network, and investigators including Paul Nutting and Larry Green built an infrastructure in which the unit of study is the practice and the sampling frame is the panel. Questions that cannot be asked anywhere else become answerable: what happens to the patient with a first episode of low back pain, how often a common presentation turns out to be serious, what the natural history of a complaint looks like when it never reaches a specialist.

Stepped-wedge cluster randomization solves the problem of withholding. Michael Hussey and James Hughes set out the design and its analysis in Contemporary Clinical Trials in 2007. Every cluster eventually receives the intervention. What gets randomized is the order in which clusters cross over, not whether they cross over at all. The Gambia Hepatitis Intervention Study, begun in the 1980s to evaluate staged rollout of infant hepatitis B vaccination, is the design's most-cited early application. A health system will not withhold a promising program from half its patients for two years. It will stage the rollout anyway, because hiring and training force staged rollout. The stepped wedge takes a constraint that already exists and converts it into a randomization schedule.

The n-of-1 trial solves the sample size problem by abandoning it. Gordon Guyatt and colleagues described the method in the New England Journal of Medicine in 1986: one patient, a chronic and stable condition, a treatment with rapid onset and offset, randomized order of active and placebo periods, blinded assessment, repeated crossovers until the pattern is clear. The unit of analysis is the person. The infrastructure required is a willing patient, a cooperating pharmacy, and a clinician who keeps the schedule. A two-physician rural practice can run this design, and the question it answers, whether this drug works in this patient, is the question primary care asks all day.

Realist evaluation changes the question. Ray Pawson and Nick Tilley argued in Realistic Evaluation (1997) that asking whether a complex program works is the wrong question when the program operates through people in a social setting. The answerable question is what works, for whom, in what circumstances, and by what mechanism. A care coordination program that succeeds in a suburban practice and fails at a border clinic has not produced contradictory findings. It has produced evidence that the mechanism depends on context, and a specification of that dependence is the useful research output.

RE-AIM makes external validity an outcome rather than a limitation paragraph. Russell Glasgow, Thomas Vogt, and Shawn Boles proposed the framework in the American Journal of Public Health in 1999: evaluate an intervention on reach, effectiveness, adoption, implementation, and maintenance. Reach and adoption determine whether anything happens at population scale. An intervention with a large effect that reaches four percent of an eligible panel does less good than a weaker one that reaches seventy percent. Efficacy trials report the first number and are silent on the second.

Qualitative description answers the question that comes before efficacy. Margarete Sandelowski's "Whatever happened to qualitative description?" in Research in Nursing & Health in 2000 defended the careful low-inference descriptive study against a methodological culture that treated it as unserious work on the way to something else. In primary care the prior question is almost always acceptability. If patients will not come, will not take it, or will not disclose the thing the intervention depends on, then an effect size measured elsewhere is a fact about a population that does not include them. Establishing that requires asking them systematically and reporting what they said.

What Nursing Already Knew About Method

Nursing's central objects never reduced cleanly to a drug trial: symptom experience, self-management, the meaning a patient assigns to a diagnosis, whether a person feels safe enough to tell the truth in a clinical encounter. The discipline had to build and adopt methods that could produce defensible knowledge about them. Madeleine Leininger's ethnonursing, the subject of Issue 019, is one of those methods. These are not soft alternatives to the trial. They are instruments for questions the trial cannot reach.

Issue 026 of this publication argued that nursing's position on those methods is doctoral and old: qualitative and mixed designs have been first-class in nursing PhD training for fifty years, and the methodological vocabulary was already in that curriculum when medicine went looking for it. The claim here is narrower and it is about fit. A practice working under constraint has the same problem nursing had. What is being studied is a person's conduct in a setting, and the setting is doing much of the causal work.

Issue 009 of this publication named the clinical form of that problem. A plan that ignores the determinants of adherence books the failure against the patient rather than against itself. The methodological form governs the study rather than the visit. Cost at the pharmacy counter, and whether the last clinician who explained the regimen was believed, are not confounders to be balanced across arms and adjusted away. They are the mechanism. A design that randomizes them into the background has thrown away the finding it was built to produce.

The consequence for an investigator working in a resource-limited practice is direct. The questions that matter most in those settings — why people do not come, what the treatment costs them in ways the chart does not record, what the clinic looks like from the waiting room — are questions nursing science has been answering with method. A physician who wants to answer them has to learn from a discipline that medical training taught him to file below his own.

The Direction of Transfer

Community-oriented primary care was invented in one of the poorest places in the world and imported to one of the richest.

Sidney Kark and Emily Kark opened the Pholela Health Centre in rural Natal, South Africa, in 1940. The method was a discipline rather than a slogan: define the community you serve, measure its health directly, act on what the measurement shows, then measure again. Kark and John Cassel reported on the work in the South African Medical Journal in 1952. Infant mortality in the served population fell over the following years, and it fell because the health centre treated malnutrition, water supply, and household conditions as clinical targets rather than as background to the clinical targets.

H. Jack Geiger went to South Africa as an American medical student in the 1950s and studied the Karks' model directly. In 1965 he and Count Gibson opened the first neighborhood health centers in the United States, at Columbia Point in Boston and then at Mound Bayou in the Mississippi Delta, funded through the Office of Economic Opportunity. Those two centers are the direct ancestors of the roughly fourteen hundred federally qualified health centers operating in this country today. The method traveled from rural South Africa to the American Delta, not the other way.

Issue 009 argued that the Diploma in Tropical Medicine teaches reasoning under diagnostic constraint, population-level clinical thinking, and the explicit treatment of culture as clinical input, and that none of the three is tropical. The same holds for method. Constraint is a design condition rather than a deficiency, and designs that work under it move upward well.

The question that gets asked about a tribal clinic, a community health center, or a rural two-physician practice is whether the site is research-capable. It is the wrong question. Those sites can run a stepped wedge, because they stage their rollouts already. They can run n-of-1 trials, because the patient is there and comes back. They can produce a denominator no referral center can produce. What they usually cannot do is meet the infrastructure requirements of a design that was never built with them in mind.

The right question is whether the method is site-capable.

What is missing is not capability. It is funding that follows the multicenter explanatory trial, promotion criteria that reward it, and a methodological hierarchy that files everything else as preliminary. Those are choices, and they are visible in the shape of the evidence base. There is a hole in it the exact size of the population most of American primary care actually sees.

The methods that would fill it are not new and they are not lesser. Most of them were built by people who had to answer the question with what was in the room.

About the author

Peter Schindler, MD, PhD is an Assistant Professor of Medicine and Associate Program Director of the Community Health Center Family Medicine Residency Program at the University of Nebraska Medical Center. He practices at Winnebago Comprehensive Health System, OneWorld Community Health Center, and Nebraska Medicine. He completed a Primary Care Research Fellowship at McGill University and holds a BSN from the University of Wisconsin-Oshkosh, an MS and PhD in nursing from Emory University's Laney Graduate School, a Diploma in Tropical Medicine from the Liverpool School of Tropical Medicine, and an MD from the Medical College of Wisconsin-Green Bay. He is board certified by the American Board of Family Medicine. The Interprofessional publishes new essays every week at the intersection of medicine, nursing, and the clinical knowledge that lives between them.

Disclaimer. The views in The Interprofessional are Peter Schindler's own and do not represent the official positions of the University of Nebraska Medical Center, Winnebago Comprehensive Health System, OneWorld Community Health Center, Nebraska Medicine, or any other affiliated institution.