Frequency distribution of health disorders in primary care—its consistency and meaning for diagnostics and nomenclature

Our objective is to reassess Braun’s finding of a consistency of frequencies by comparing case statistics from several practices as well as different observation periods. We hypothesized a significant correlation in frequency ranking between surveys, particularly between those that are close in time.

Braun first started with a visual presentation of his year-over-year frequency data and then compared them with data from the UK [11]. There was an obvious similarity as there was in our material (Fig. 4). Early morbidity surveys in Canada by Lynn Curry and Karen MacIntyre also described “similarity between the morbidity patterns in all areas”. They already used a statistical method for their comparison: Frequency ranks in 15 profile studies in family practice showed similarity with Kendall’s coefficient of concordance of 0.65 [4]. Later Braun and Haber used Spearman’s correlation analysis and had coefficients of 0.47–0.71 [24]. It again confirmed the similarity. Our study found positive Spearman’s correlation coefficients between 0.30 and 0.82 when all different practices were compared (Table 3 correlation matrix). The results support our hypothesis of a general inherent similarity. If variations occur they are due, on the one hand, to the different survey periods, on the other hand, to differences in nomenclature.

Braun himself observed that the ranking was influenced by changing living conditions [25]. Our comparison over five decades equally reflects these changes. We see the well-known epidemiologic changes in diseases of affluence like diabetes and hypertension. Likewise, life expectancy has risen, which also influences the types of disorders that are presented to primary care practitioners. Furthermore, the increase of recorded coronary heart disease, osteoporosis, atrial fibrillation or depression may be due to better diagnostic procedures and therapeutic options. The high figures of hyperuricemia, of lipid or thyroid disorders in all recent surveys must not only be interpreted as a change in the morbidity spectrum, but could be the result of lowered thresholds, or more frequent testing or better monitoring of chronic conditions in electronic medical records (Supplementary_material_2_Table_256_ranks_english).

Changes within the healthcare system may play a role in the case distribution of health disorders seen by primary care physicians. During our observation period, the rise of private medicine had only just begun, so there was no noticeable impact. From Prosenc’s observation we know that an increase in the number of practicing specialists in his town, in the sixties, led to an overall decrease in cases in his primary care practice [26]. However, in contrast to Prosenc’s findings Fink’s practice—comparing observation periods in the nineties with 2005–2009—showed an increase in both patients and the number of health problems per patient (Table 2). A second point of consideration is that out-of-hours services in our hospital outpatient departments have increased in recent years. Patients may or may not be sent back to the practitioner. No doubt, it would be useful to monitor practice morbidity in sentinel practices, in different medical fields and institutions, and to assess the influence of changes in the healthcare system on the case distribution.

Based on the expected consistency of illness frequencies especially between surveys close in time, we looked for the relationship between the labeling of health disorders and the similarity of case frequency distributions.

The rankings of illnesses in the different practices in the eighties and nineties, where primarily casugraphic labels were used, are highly consistent (up to 0.82), whereas the most recent data, where the physicians’ labels were harmonized retrospectively, showed less strong correlations with each other (Table 3). But, as visualized in the heatmap (Fig. 2), the comparison of consultation results year by year within the same practices yielded a very strong correlation with a Spearman correlation coefficient of up to 0.96 (Supplementary_material_3_corOverallyearly and Fig. 4). Similarly, Crombie et al. had observed a “consistency of any individual doctor’s pattern of diagnostic recording from one year to another” [27]. A closer look at the respective first 10–20 ranks in our yearly comparison reveals a possible source of disparities (Fig. 3): In the process of harmonizing the nomenclature for the analyses, identical clinical expressions were analyzed as they were recorded. As shown in Fig. 3, we found that some ranks differ considerably among practitioners but are very consistent every year within the same practice (e.g. myalgia, acute bronchitis, cough, strep throat, dizziness). These results suggest that clinical terms were used with a variable individual meaning. In the transnational study by Jean Karl Soler and collaborators, using International Classification of Primary care (ICPC-2) codes, the reasons for encounter (RfE) codes showed, “striking similarities in the incidence or prevalence rates”, but considerable variability in the consultation results, coded as “episode of care” (EoC) [5]. As seen in many studies, similar ranks are observed especially when clear medical terms are available or when clinical conditions are evident (e.g. excessive earwax, certain skin diseases) [3, 5, 28]. The better a clinical condition can be assessed by general practitioners, e.g. hypertension, diabetes, low back pain, the more consistency is found with other practices. When dealing with nonspecific symptoms however, physicians develop their own specific ways of assessing conditions. They seem to have different preferred labels and different codes when no firm diagnosis is reached [7]. There have been several attempts to subsume primary care doctors’ colloquial clinical terms under appropriate codes that are ideally compatible with ICD-codes. It started in the sixties with the Royal College of General Practitioners’ Classification, the US Ambulatory Medical Care Classification of Symptoms (NAMCS), and then early versions of the ICPC, the International Classification of Health Problems of Primary care (ICHPPC) [29]. This was followed by Oscar Rosowsky’s efforts to integrate Casugraphic labels into ICD, as well as to assist practitioners in their diagnostic considerations [30].

When there is variance in morbidity figures, it seems possible that there is variance in the diagnostic approach too [31]. Braun described his research experience: “The entries must be made more precisely, without violating the facts. For this purpose, it has been found to be of particular importance to delimit each individual case or each individual rubric from all practically important diagnostically related cases and rubrics. … The way to a usable classification is inextricably linked to a realistically developed and sophisticated practical diagnostics and vice versa. Diagnostics and classification represent a unit, like columns of fluid in communicating vessels” (Braun) [11, p. 60, 61¹]. Diagnostic reasoning influences classification and vice versa [32].

Beyond epidemiological issues, the case distribution provides insights into the everyday challenges of a primary care physician. The comprehensive approach taught in medical school often cannot be followed due to known constraints and limitations in frontline medical care. “A very high percentage of cases … will not be diagnosed in the accepted sense of the word, but can only be classified according to the leading symptom or symptom complexes. Many of these will be minor illnesses, clearing up after a short course. … Yet among this mass of clinical material there will occur, rarely, but regularly, potentially dangerous conditions” [33]. The power law morphology of illness ranking reminds that diagnostic considerations have to do with risk management. In the so called “fat tail” the least certain diagnostic units prevail, like fever, respiratory symptoms, pain (muscular-skeletal, abdominal or precordial), headache, dizziness (Fig. 4). Herein lies a high risk of a hidden life threatening condition or “black swan” event, as they are called by today’s scientific forecasting [34]. The physician must always be prepared for such a rare event, despite its low probability. An unwarranted disease label here is a risk for premature closure [35‐37]. As diagnoses are traditionally expected, physicians may hesitate to classify on the symptom level [38]:

This again leads to nomenclature issues. “… new codes to denote diagnostic uncertainty in the patient-provider encounter” [39] are needed. Knowledge of the “law of case distribution” suggests that we can expect a manageable number of 200–300 health problems without a confirmed diagnosis to occur regularly. The casugraphic diagnostic rubrics emerged as a by-product of Braun’s case statistic [40]. They were defined by delimiting each rubric from related rubrics and by listing the most important potentially dangerous conditions to be considered in the differential diagnostic procedure [23, 41].

It is conceivable to integrate these casugraphic units as prototypes of consultation results into automatic coding software. Kazem Sadegh-Zadeh, author of the “Handbook of Analytic Philosophy of Medicine” [42] considered Braun’s approach to be more or less the only approach suitable for unambiguous and automatic data assignment (M. Konitzer, personal communication, 2012). Embedded in a problem-oriented electronic patient file, already being developed by Wolfgang Edinger, it could be a helpful tool both for the diagnostic work with the patient and for coding [43, 44]. Undifferentiated symptoms and syndromes become as manageable as if they were diagnoses, but with the reminder of an open situation, where attentive observation is required.

Data comes from everyday practice and was collected very conscientiously and carefully by physicians, covering the entire spectrum of diseases and an extremely long period of time. A limitation is that we had to rely on relatively old published data when electronic medical records did not exist. The fact that the recorded diagnoses were not influenced by financial incentives, as observed in current health insurance cost analyses [45, 46], can be seen as a strength.

General practices were not sampled at random, but all of them operated within an unselected population and within the national health care system. The patient structure is only partially known or was not recorded because the survey was limited to recording health disorders. A population-based survey with precise age and gender distribution was not intended; the focus was on the physician’s recording of the episode of illness he or she was consulted for.