Ecology of care: the disease primary care sees

A student finishes the hospital internship convinced that chest pain almost always hides something serious. She spent months seeing heart attack, dissection and embolism on the ward. In her first month at a primary care service, she sees several people with chest pain every week, and almost all of them have muscle pain, reflux or anxiety. She did not become less attentive. The population is what changed.

This phenomenon has a name: the ecology of care, the study of who gets sick, who seeks help and where that help happens. This text presents the classic square that gave rise to the concept, shows what updates in different health systems found and explains why the care setting changes the frequency of diseases, the spectrum of severity and the performance of tests. At the end, it discusses what this changes in the reasoning of those who work at the entry point of the system.

Key points

• Most people who have some health problem in a month do not consult, and a very small share reaches the hospital.
• This pattern repeats in studies from several countries over almost a century, with differences linked to the organization of each system.
• Those seen in primary care have, on average, a lower probability of serious disease than those who arrive at the hospital.
• The sensitivity and specificity of a test can change depending on the population in which it is used, because of the spectrum of the disease.
• Alarm symptoms have low predictive value in primary care, but not negligible, and this value rises with age.
• Many common symptoms improve with time, and some remain without a medical explanation even after years.
• Bringing the reasoning learned in the hospital to primary care tends to generate too many tests and overdiagnosis.

Summary

• The White square: who gets sick, who seeks help, who reaches the hospital
• What changed and what repeats over almost a century
• What changes from one level to another
• Spectrum: the same test in another population
• Chest pain: an example of a setting in practice
• Alarm symptoms in primary care
• Common symptoms and what happens to them
• What this changes in the clinician's reasoning
• Frequently asked questions
• Final checklist

The White square: who gets sick, who seeks help, who reaches the hospital

In 1961, White and colleagues published a figure that became classic in health education. From household surveys and service records, they estimated what happens to a thousand adults over a typical month. The historical review by Spinelli and colleagues reproduces the numbers: of every thousand adults, about 750 report some illness or injury, 250 consult a doctor, nine are admitted, five are referred to another doctor and one reaches a university medical center (Spinelli, 2023).

The figure is usually drawn as a large square, which represents the population, with ever smaller squares inside it. The smallest of them, the university hospital, is almost a dot. The authors' message was direct: the experience of those who train and do research inside that dot does not represent disease as it exists in the community.

Some details of the design matter for reading the numbers carefully. The data came from surveys done in two countries, in different periods, and excluded people under 16 and uncomplicated pregnancies (Spinelli, 2023). The unit of time was the month, not the year. This means the same person can be in the population square one month and in the hospital one the next.

The value of the square is not in the precision of the numbers, which vary greatly depending on place and time. It is in the format. It forces you to think about where the patients we study come from and who is left out when we look only at the hospital. For clinical reasoning, this question has practical consequences, which appear in the following sections.

Three ideas sum up what the square teaches. The first is that most illness is resolved at home, without contact with the health system. The second is that, among those who seek help, most are seen and resolved outside the hospital. The third is that the university hospital, where most professionals train and most research is done, sees a minimal and highly selected fraction of the population.

What changed and what repeats over almost a century

Forty years later, Green and colleagues redid the exercise with data from a 1996 national survey. In the new version, out of every thousand people in a month, 800 have symptoms, 327 think about seeking care, 217 go to a doctor's office, 113 to a primary care service, 65 to complementary practices, 21 to a hospital outpatient clinic, 14 receive home care, 13 go to the emergency department, eight are admitted and one reaches an academic center (Spinelli, 2023). The update included children and new care settings, but the overall format stayed the same.

The review by Spinelli and colleagues gathered the ecology of care studies published between 1928 and 2018, in eight countries, and found two regularities over 90 years: a large proportion of people with health problems who do not consult and a low number of admissions and referrals (Spinelli, 2023). The authors also call attention to common limitations: the studies use different definitions of symptom and of consultation and almost always depend on what people remember and report.

The differences between the studies say as much as the similarities. An analysis of the same 1996 survey showed that the doctor's office was the most common care setting in all subgroups, and that not having a usual source of care was the only characteristic independently associated with less care in all five settings studied (Fryer, 2003). Having health insurance and a usual source of care at the same time was associated with more care in all settings, except the emergency department.

More recent studies show that the pattern resists policy changes. An analysis of national surveys between 2002 and 2016, with 525,804 person-years, found a decline in contact with primary care physicians, steeper among older adults and people in poor health, and almost no change in the overall picture after a major coverage reform (Johansen, 2019). A provincial cohort, which followed adults from 2002 to 2017, found nearly stable patterns despite five broad system changes, with admission to an academic hospital at around one in every hundred adults per year (McAlister, 2020).

The organization of the system, however, changes the shape of the square. A telephone survey with 3,508 interviews, compared with the international literature, found very high use of specialists and of hospital admission in a system with free access to all levels and no clear boundary between primary and secondary care (Hoffmann, 2019). The authors attribute part of this use to the lack of the coordinating role of a strong primary care.

What changes from one level to another

The square shows how many people reach each level. The clinical question is different: are these people different? The literature's answer is yes, and systematically so. Those who reach the hospital passed through a series of filters. They had symptoms persistent or severe enough to seek help, were assessed by someone who decided to refer them and, often, had already undergone tests that reinforced the suspicion.

Starfield and colleagues, in a review on the contribution of primary care to health systems, sum up the consequence: specialists trained in the hospital see patients who do not represent the way problems present in the community, and community patients have a much lower prior probability of serious disease (Starfield, 2005). Because the properties of tests depend on that probability, those who carry the hospital's expectation into the community tend to overestimate the chance of disease.

The same review points out that first contact with primary care is associated with more appropriate and lower-cost care, and that the use of tests and treatments guided by an overestimated probability brings adverse effects for patients (Starfield, 2005). The difference is not one of competence. An excellent specialist may err more in the community precisely because they calibrated their own eye on a selected population.

The table below sums up the typical differences between the two settings. They are average tendencies, not rules for each person.

Characteristic	Primary care	Hospital or referral service	What changes in the reasoning
Who arrives	People from the community, many at the start of symptoms	People filtered by severity, persistence or referral	The starting probability is different
Frequency of serious disease	Low for most complaints	High for the complaints that led to admission	The same symptom carries different weight
Stage of the disease	Early, with vague and incomplete signs	Advanced or already defined	Classic findings appear less often
Problems per person	Several at once, physical and psychosocial	Usually one main problem	The complaint must be read in context
Test performance	May differ from what hospital studies describe	Close to that of studies done in the same setting	You need to know where the test was validated
Role of time	Return visits and observation are viable tools	Faster decisions and more tests	Waiting with a plan can be an option

Formal reasoning about pre-test probability and base rate is explained in other blog posts, with the example of the same test in different populations. The point here comes before the calculation: knowing which square of the figure the person is in.

Spectrum: the same test in another population

It is often taught that sensitivity and specificity are fixed properties of a test, while predictive values change with prevalence. The first part of that statement is a simplification. Leeflang and colleagues analyzed 23 meta-analyses of diagnostic accuracy, with 416 studies in total, and found variations from zero to 40 percentage points in sensitivity or specificity between the studies with the lowest and the highest prevalence (Leeflang, 2013). The association with prevalence was statistically significant in eight of those meta-analyses.

The most frequent direction was lower specificity when prevalence was higher, with no systematic effect on sensitivity (Leeflang, 2013). The authors propose patient spectrum as the likely mechanism, that is, differences in disease severity and in similar conditions between the populations studied. The practical recommendation is to use prevalence as a clue for choosing the accuracy studies that most resemble your own setting.

Usher-Smith and colleagues describe this spectrum effect in diagnostic, screening and risk prediction tests (Usher-Smith, 2016). A test developed in a population with high prevalence tends to have lower sensitivity and higher specificity when used in a population with low prevalence. More severe disease is easier to detect, which makes the test look better in the populations where it predominates.

Study design also matters. A study that compares patients with confirmed disease in hospital with healthy controls from the community brings together the two extremes of the spectrum and tends to overestimate accuracy (Usher-Smith, 2016). The example given by the authors is simple: a pregnancy test evaluated in a prenatal clinic will not perform the same way when used in people from the community who do not yet know whether they are pregnant.

For those working in primary care, this raises three questions before trusting an accuracy figure:

• Where was the study done, and who were the patients included?
• Did the patients have the same stage and severity of disease that I usually see?
• Did the group without disease include people with other conditions that mimic the disease being sought?

The post on how to read diagnostic studies details these points, and the text on sensitivity, specificity and predictive values shows how they combine with the starting probability.

Chest pain: an example of a setting in practice

Chest pain is a good example because it is alarming in any setting. In the emergency department, the dominant concern is acute coronary syndrome, and tools such as the HEART score were designed for that population. In primary care, the distribution of causes is different.

A systematic review with meta-analysis gathered 11 studies done in primary care, with about 6,500 patients with chest pain (Haasenritter, 2015). Chest wall syndrome accounted for a quarter to half of the cases, respiratory and psychogenic causes for about one in ten to nearly one in five each, and stable coronary disease for about seven to 11 in every hundred (Haasenritter, 2015). Acute coronary syndrome or infarction ranged from one to four in every hundred patients (Haasenritter, 2015).

The authors present these numbers as pretest probabilities useful to the family physician. They do not say that chest pain in primary care is benign. They say that the starting point differs from the emergency department, and that assessment must take this into account in both directions.

In practice, this translates into a few stances. The first is to recognize that the most likely cause of chest pain in primary care is usually musculoskeletal, without ruling out the serious possibility. The second is to use history and examination to identify who is in the small high-risk group and needs urgent assessment. The third is to avoid applying to patients in the office tools and thresholds created for the emergency department without knowing whether they have been validated in primary care. And, when the initial assessment shows no severity, to arrange follow-up and guidance on warning signs.

The clinical history as the first tool is the main resource for this triage, especially where tests are not available on the spot.

Alarm symptoms in primary care

Alarm symptoms are the other side of the same coin. Taught in the hospital as signs of serious disease, they appear frequently in primary care in people who do not have cancer. Jones and colleagues used a large general practice database with 762,325 patients from 128 practices to measure the positive predictive value of alarm symptoms recorded for the first time (Jones, 2007).

Over three years, the predictive value of hematuria for urinary cancer was about seven in every hundred men and just over three in every hundred women (Jones, 2007). Rectal bleeding reached about two in every hundred patients for colorectal cancer (Jones, 2007). The values increased with age and were higher in men; hemoptysis in men aged 75 to 84 reached about one in six (Jones, 2007).

These numbers are low when compared with the impression of those who learned the symptoms in the hospital, where people with hemoptysis often arrived with an abnormal image already. At the same time, they are high enough to justify investigation in many situations, especially in older groups. What the study shows is that the value of a symptom depends on who presents it and where it is assessed.

For reasoning, this means that an alarm symptom in primary care does not amount to a diagnosis, but it changes the probability substantially. Age, sex and symptom duration adjust this weight. The first recording of the symptom is a decisive moment for deciding to investigate, monitor or refer, and a follow-up plan is part of the response when the initial investigation is negative.

Common symptoms and what happens to them

Most consultations in primary care begin with a physical symptom, and a good part of these symptoms does not lead to a defined diagnosis. Jackson and Passamonti followed 500 consecutive patients who sought an internal medicine outpatient clinic with a physical complaint and reassessed them at several stages up to five years later (Jackson, 2005).

In five years, about four in five patients reported improvement, a little more than half had the symptom resolved, and about a third of the symptoms remained without medical explanation (Jackson, 2005). Most people with unexplained symptoms did not have a mental disorder, and mood or anxiety disorders were associated neither with the lack of explanation nor with a worse course. Worse function, a symptom of longer duration at the first visit, worry about illness, and absence of resolution at three months were associated with a lower chance of improvement (Jackson, 2005).

These findings have two consequences. The first is that time is an ally of diagnosis in primary care: many symptoms resolve, and those that persist or change character become more informative. The second is that a symptom without explanation is not synonymous with a psychological cause, and labeling the person that way can end too early the investigation of something that may still appear. Diagnostic uncertainty is therefore a normal condition of this setting, not a sign of failure.

What this changes in the clinician's reasoning

Bringing reasoning calibrated in the hospital to the point of entry has a cost. Tests ordered to rule out unlikely diseases produce falsely positive results, further tests, and sometimes treatment of findings that would never have caused a problem. The text on how the harm of overdiagnosis reaches the patient describes this cascade, and the post on low-value tests and practices shows how to recognize them.

The opposite error also exists. Primary care sees so many benign cases that the eye can grow accustomed and let the rare one pass. Classic heuristics, such as availability and representativeness, work well most of the time and fail precisely in atypical presentations. The ecology of care does not authorize lowering one's guard; it helps decide where to place it.

The same reasoning applies to digital tools. Symptom checkers and algorithms trained on hospital data or on vignettes may perform differently when used with real flesh-and-blood patients in the community. The question of where the tool was validated is the same one asked of any test.

Some practices help apply the ecology of care day to day:

• Ask, for each accuracy or risk figure, in which population it was measured.
• Estimate the starting probability from the setting and the person, before ordering tests.
• Separate the few signs that require immediate action from the many that allow observation.
• Use the scheduled follow-up as part of the reasoning, with clear criteria for reassessment.
• Record in the chart what was considered and why it was decided to investigate or wait.

Frequently asked questions

What is the ecology of care?

It is the study of how people in a population become ill, seek help, and use the different levels of the health system over a period. The classic model shows that most problems never reach the hospital (Spinelli, 2023).

Do the numbers of White's square still hold?

The exact numbers vary with place, time, and how the system is organized. The general shape, with many people with problems, fewer consulting, and few hospitalized, repeats across studies spanning almost a century (Spinelli, 2023).

Are sensitivity and specificity not fixed?

Not always. In several meta-analyses, they varied with prevalence, probably because of differences in the disease spectrum among the populations studied (Leeflang, 2013). So it is worth looking for studies done in a setting similar to yours.

Is chest pain in primary care almost always benign?

The most common causes are musculoskeletal, respiratory, and psychogenic, and acute coronary syndrome is infrequent in this setting (Haasenritter, 2015). Infrequent is not absent, and the initial assessment serves precisely to find those who need urgent care.

Does a symptom without explanation mean an emotional cause?

Not necessarily. In a five-year follow-up, most people with symptoms without medical explanation did not have a mental disorder (Jackson, 2005).

Does this text guide individual management?

No. It explains a concept from primary care and the evidence about it, and it does not replace the clinical assessment of each patient.

Final checklist

• Remember at which point of White's square the person in front of you is.
• Estimate the starting probability based on the setting, before thinking about tests.
• Check where the test, score or tool you intend to use was validated.
• Be wary of accuracy from studies that compare hospital patients with healthy people from the community.
• Treat alarm symptoms as a change in probability, adjusted for age and sex.
• Use the scheduled follow-up and the guidance on warning signs as part of the plan.
• Do not confuse a symptom without explanation with an emotional cause.
• Record the reasoning and the criteria for reassessment.

References

1. Fryer GE Jr, Green LA, Dovey SM, Yawn BP, Phillips RL, Lanier D. Variation in the ecology of medical care. Ann Fam Med. 2003. PMID 15040437. DOI 10.1370/afm.52.
2. Haasenritter J, Biroga T, Keunecke C, et al. Causes of chest pain in primary care: a systematic review and meta-analysis. Croat Med J. 2015. PMID 26526879. DOI 10.3325/cmj.2015.56.422.
3. Hoffmann K, Ristl R, George A, Maier M, Pichlhöfer O. The ecology of medical care: access points to the health care system in Austria and other developed countries. Scand J Prim Health Care. 2019. PMID 31512566. DOI 10.1080/02813432.2019.1663593.
4. Jackson JL, Passamonti M. The outcomes among patients presenting in primary care with a physical symptom at 5 years. J Gen Intern Med. 2005. PMID 16307629. DOI 10.1111/j.1525-1497.2005.0241.x.
5. Johansen ME, Richardson CR. The ecology of medical care before and after the Affordable Care Act: trends from 2002 to 2016. Ann Fam Med. 2019. PMID 31712291. DOI 10.1370/afm.2462.
6. Jones R, Latinovic R, Charlton J, Gulliford MC. Alarm symptoms in early diagnosis of cancer in primary care: cohort study using General Practice Research Database. BMJ. 2007. PMID 17493982. DOI 10.1136/bmj.39171.637106.AE.
7. Leeflang MM, Rutjes AW, Reitsma JB, Hooft L, Bossuyt PM. Variation of a test's sensitivity and specificity with disease prevalence. CMAJ. 2013. PMID 23798453. DOI 10.1503/cmaj.121286.
8. McAlister FA, Tonelli M, Wiebe N, et al. The ecology of medical care for adults in Alberta, 2002/03 to 2016/17: a retrospective cohort study. CMAJ Open. 2020. PMID 32184280. DOI 10.9778/cmajo.20190188.
9. Spinelli H, Trotta A, Martinovich V, Alazraqui M. Studies on the ecology of medical care: a comparative analysis in historical perspective, 1928-2018. Salud Colect. 2023. PMID 38006657. DOI 10.18294/sc.2023.4549.
10. Starfield B, Shi L, Macinko J. Contribution of primary care to health systems and health. Milbank Q. 2005. PMID 16202000. DOI 10.1111/j.1468-0009.2005.00409.x.
11. Usher-Smith JA, Sharp SJ, Griffin SJ. The spectrum effect in tests for risk prediction, screening, and diagnosis. BMJ. 2016. PMID 27334281. DOI 10.1136/bmj.i3139.