OUP user menu

Do specialized centers and specialists produce better outcomes for patients with chronic diseases than primary care generalists? A systematic review

Piet N. Post, Jolanda Wittenberg, Jako S. Burgers
DOI: http://dx.doi.org/10.1093/intqhc/mzp039 387-396 First published online: 4 September 2009


Purpose Although specialized centers are generally accepted for treatment of relatively uncommon diseases, such as cystic fibrosis, statements regarding the amount of expertise or minimum number of patients treated are increasingly included in guidelines for the treatment of other chronic diseases such as rheumatoid arthritis and diabetes mellitus.

Data sources Medline and Embase from 1987 through March 2008 were searched.

Study selection Studies reporting the effect of treatment in a specialized or high-volume center or by subspecialists on a clinically relevant outcome.

Data extraction Two reviewers extracted the data independently and assessed the methodological quality.

Results of data synthesis We included 22 articles. Two randomized-controlled trials and a quasi-experimental study compared the effect of outpatient team care with traditional outpatient care for patients with rheumatoid arthritis. These studies showed no difference or were inconsistent. Studies on the outcomes of care for diabetic patients (5 prospective or historical cohort studies and 10 retrospective cohort studies) were generally of poor quality. Studies comparing the subspecialist care with the care provided by general internists or primary care providers produced inconsistent results. Similar inconsistency and poor quality were found for three observational studies on cystic fibrosis.

Conclusion The available literature suggests that among patients with rheumatoid arthritis, diabetes mellitus or cystic fibrosis, outcomes are not superior in specialized centers or with subspecialists compared with other forms of chronic illness care.

  • quality indicators
  • patient outcomes (health status, quality of life, mortality)
  • case-mix or risk adjustment
  • health policy
  • guidelines


Over the past decade, the need for information about the quality of care services has increased. Information on treatment outcomes is usually preferred. However, this type of information is often unavailable and may not always be valid for use as a quality indicator [1]. Instead, process measures of quality, derived from clinical practice guidelines, are widely adopted as indicators of quality [2]. Furthermore, the annual volume of patients treated in a hospital is increasingly used as an indicator of quality. The use of this indicator emerged after Luft et al. [3] showed that hospitals performing more than 200 procedures annually had substantially lower surgical mortality rates for some surgical procedures than lower volume hospitals.

After the volume–outcome relationship was confirmed in more recent studies, the use of volume as surrogate indicator of quality has been promoted by the Leapfrog Group and others [4]. The majority of studies included patients undergoing invasive procedures whose outcomes can be observed shortly after the intervention and are relatively easy to measure. For non-invasive procedures, however, the relationship between volume and outcome is unclear [5]. Yet, statements regarding the amount of expertise and/or the minimum number of patients treated do appear in guidelines concerning other chronic diseases [6, 7]. Although several reasons exist to recommend treatment in a specialized center, these statements are usually justified by claiming that specialized centers produce better outcomes.

We wondered whether the available evidence indeed supports better outcomes in specialized centers. Therefore, we set out to review the evidence for a disease with general consensus that treatment should take place (by subspecialists) in a specialized center (cystic fibrosis) as opposed to two chronic diseases for which this is less clear (rheumatoid arthritis and diabetes mellitus). Apart from the value of specialized care, we also reviewed the evidence concerning the value of treatment in a high-volume center. Although specialization and high volume are not synonymous, similar processes may determine their relation with outcome [8]. We were interested whether treatment in a specialized (or high-volume) center (or by subspecialists) results in better outcomes for patients, not whether it results in better process measures. Therefore, we conducted a systematic review of the literature to summarize the evidence on the association between healthcare provision in a specialized center or by subspecialists and the outcome of treatment for these three chronic diseases.


Data sources and searches

An experienced librarian searched the databases Medline and Embase, using Winspirs software (Silverplatter Information Inc, Norwood, MA) from 1987 through 2008 combining search terms defining treatment in a specialized center or by subspecialists (professional-competence, caseload, workload, volume, number, frequency, utilization, expertise) with search terms defining outcomes (mortality, morbidity, treatment-outcome, outcome-assessment-healthcare, outcome-and-process-assessment-health-care, quality-of-health-care, program-evaluation, peer-review-healthcare, medical-audit, delivery-of-healthcare-integrated, health-services-accessibility, product-line-management, delivery-of-healthcare, statistics-and-numerical-data or epidemiology) and the disease of interest (arthritis-rheumatoid, juvenile rheumatoid arthritis, rheumatoid-nodule, Sjogren's syndrome, arthritis, polyarthritis, rheumatic diseases; cystic-fibrosis, mucoviscidosis, Pseudomonas aeroginosa, cf.; diabetes-mellitus, diabetes-mellitus-Type-1, diabetes-mellitus-Type-2, diabetes-gestational, hemoglobin-A-glycosylated).

Study selection

Articles were selected if volume, centralized treatment or subspecialist treatment was an independent variable and its impact on a relevant clinical outcome of cystic fibrosis, rheumatoid arthritis or diabetes mellitus was recorded. Articles reporting results on process measures were not included. Articles were also excluded if they did not contain primary data. If various articles were based on the same dataset, the most recent or most informative article was included. Although randomized studies are preferred in the study of intended effects [9], we included both randomized and observational studies to enhance the yield of our search. Similarly, we did not exclude articles on the basis of the quality assessment, but discuss the implications of the observed deficits in study design on our results.

Data extraction and quality assessment

Two of the authors (P.N.P. and J.W.) extracted the data independently and assessed the quality of the selected papers, using a quality assessment tool developed by Thomas [10]. This instrument has been validated for both randomized and non-randomized studies [11]. Using this instrument, the following items were systematically examined and scored as strong, moderate or weak according to the explicit criteria: selection bias, study design, confounders, blinding, data collection methods and withdrawals and dropouts. Although the instrument allows for an overall score, we focussed in particular on study design (preventing confounding by indication) and confounders (in particular in non-experimental or quasi-experimental studies). Case-mix adjustment was considered adequate if results were adjusted for differences in case mix at baseline (before treatment). The results of data extraction and quality assessment were compared between the two reviewers. Disagreements were mainly caused by different interpretations of unclear descriptions in the articles and were resolved by discussion between the two reviewers.

Data synthesis and analysis

Because reported outcomes varied to a large extent between studies, no attempt was made to pool the results statistically. All results on clinically relevant outcomes were listed. In addition, we classified studies as positive, negative and combined those showing no significant effect or conflicting results. Only statistically significant findings (two-sided P-value < 0.05) were classified as positive or negative.


The literature search initially yielded 1119 potential articles. Of these, 110 were potentially relevant for rheumatoid arthritis, 85 for cystic fibrosis and 272 for diabetes mellitus. After selection on the basis of title/abstract and checking for cross-references, 81 articles were closely scrutinized. A total of 22 articles met our selection criteria (Fig. 1), only two of which were randomized-controlled trials (RCT). Most excluded studies addressing the value of specialized treatment did not report results on relevant outcomes but on process measures. The quality of the included studies was generally limited (Table 1). Most studies were observational studies that failed to adequately control or correct for confounding. A large variation was observed in the definition of a specialized center and in the comparison group. Only one study addressed the value of high volume vs. low volume of care, while most addressed the value of some kind of specialized care compared with usual care. Among these, some studies compared subspecialty care with care by general internists, whereas other compared it with primary care by general practitioners.

Figure 1

Flow diagram of literature selection.

View this table:
Table 1

Characteristics of the included studies evaluating the effect of treatment in a high-volume or specialized center

Author (reference)YearCountryDesignCase-mix adjustmentaOverall qualityPopulation/settingn
Cystic fibrosis
 Mahadeva et al. [21]1998UKRetrospectiveWeakWeakAdult cystic fibrosis patients97
 Nielsen et al. [22]1988DenmarkHistorical cohortModerateModerateCystic fibrosis patients with explicit criteria514
 Walters et al. [20]1994UKCross‐sectional designWeakWeakAdult cystic fibrosis patients669
Rheumatoid arthritis
 Ahlmen et al. [12]b1988SwedenRCTStrongStrongOutpatient rheumatoid arthritis patients59
 Newman and Silman [15]1996UKCross-sectional designWeakWeakReumathoid arthritis patients149
 Raspe et al. [13]b1992GermanyQuasi-experimental studyStrongModerateNew referrals to tertiary rheumatoid arthritis center179
 Schned et al. [14]b1995USARCTStrongModerateChronic inflammatory arthritis118
Diabetes mellitus
 Baumer et al. [31]1997UKRetrospectiveWeakcWeakChildren with diabetes801
 De Berardis et al. [23]2004ItalyProspective cohortWeakcWeakType 2 diabetes3437
 Greenfield et al. [32]1995USARetrospective cohortWeakcWeakPatients with diabetes and hypertension1296
 Greenfield et al. [33]2002USARetrospectiveWeakcWeakAdult diabetic patients1750
 Ismail et al. [34]b2006UKProspective studyStrongModerateAmbulant patients1550
 Levetan et al. [16]b1999USARetrospective cohortStrongModerateKetoacidosis patients260
 McAlister et al. [18]b2007CanadaRetrospective cohortStrongModerateAmbulatory patients24232
 Pellegrini et al. [35]2003ItalyProspective cohortWeakcWeakDiabetic patients with hypertension1748
 Schiel et al. [36]1997GermanyHistorical cohortWeakcModerateDiabetic patients244
 Shah et al. [37]2005CanadaRetrospective studyWeakcWeakDiabetic patients4507
 Sone et al. [38]2006JapanProspective studyWeakcModerateType 2 diabetes754
 Turchin et al. [19]2007USARetrospective cohort studyWeakModerateDiabetic patients treated by primary care physicians7120
 Uchigata et al. [17]2004JapanRetrospective cohortWeakcModeratePatients with Type 1 diabetes1430
 Zgibor et al. [39]2002USAProspective cohortWeakcModerateChildhood onset diabetes429
 Zoppini et al. [40]1999ItalyRetrospective studyWeakcWeakPatients with Type 2 diabetes7148
  • aIf less than 60% of relevant confounders were controlled, this item is scored as weak in the quality assessment tool. It is scored as moderate if 60–79% were controlled and strong if 80–100% were controlled [10]. bDenotes a study of higher quality. cCase-mix data were not collected at baseline.

Rheumatoid arthritis

Two RCTs and one well-designed quasi-experimental study compared the effect of outpatient team care with traditional outpatient care (Table 2). Team care typically consisted of care by a rheumatologist, a nurse, a physiotherapist, an occupational therapist and a social worker. Care in the control group was given by a rheumatologist (and other professionals only on initiation of this rheumatologist) [12], by a physician in training [13] or by a general practitioner or rheumatologist [14]. Ahlmen et al. [12] reported an improvement in overall health, measured with the Sickness Impact Profile, in patients treated with team care, but no effect on disease activity or joint function. Raspe et al. [13] reported larger improvements in the number of swollen joints with comprehensive outpatient care but no difference in overall health. The other RCT revealed no differences on any of the outcomes measured (Table 2) [14]. The cross-sectional study by Newman et al. [15] addressed the value of treatment in a specialist clinic, but did not detect a difference in the number of swollen joints.

View this table:
Table 2

Results of studies evaluating the effect of treatment of rheumatoid arthritis in a high-volume or specialized center

AuthorInterventionControlOutcomeFollow-upResults center vs. controlOverall result
Ahlmen et al. [12]a
Outpatient team care (rheumatologists, nurse, physiotherapist, occupational therapist, social worker)Non-team care (rheumatologists and other health professional on initiation of rheumatologist)Change in overall health (SIP)1 yearIncreased to 3.6% vs. 0.1% (P < 0.05)+/=
Disease activity (Ritchie index)10.6 (SD 8.4) vs. 11.1 (SD 7.5)
Newman and Silman [15]Specialist clinic, as assessed by questionnaireGeneral or no careNumber of swollen jointsn.a.5 vs. 7 (P > 0.05)=
Raspe et al. [13]
Comprehensive outpatient care by two rheumatologists and other health professionalsTraditional care by single physician in training, assisted by one nurseNumber of swollen joints2 yearsDecreased to 10 ≥ 5 vs. 7 ≥ 3 (P < 0.01)+/=
Rating scale5.2 vs. 5.2
Schned et al. [14]aOutpatient team care: rheumatologist, education and management program with emphasis on self-managementTraditional outpatient care by primary care physician or rheumatologist
Ritchie index1 year2.0 vs. 1.2 (P = 0.5)=
ACR functional class1.7 vs. 1.7
  • SIP, sickness impact profile; SD, standard deviation; ACR, American College of Rheumatologists. ‘+’, intervention better; ‘−’, control better; ‘±’, no significant effect or inconsistent results).

  • aDenotes a study of higher quality.

Diabetes mellitus

A total of 15 studies were included, 6 of which reported on the value of subspecialist care compared with care provided by a general internist, 5 compared specialist care with primary care, 2 compared the outcomes of patients seen 1 or more by an endocrinologists with the outcomes of those who were not and 1 investigated the effect of case volume. Reported outcomes were mortality rates, diabetic complications and hypertension, while HbA1C (n = 8) was the most frequently used outcome (Table 3). Many studies were started some time after the patient or his physician had chosen for specialized or standard treatment. Although most of them seemed to correct for differences in case mix, the differences in case mix were frequently not assessed at baseline, but years after the patient or his physician selected the type of care (Table 1).

View this table:
Table 3

Results of studies evaluating the effect of treatment of diabetes mellitus in a high-volume or specialized center

AuthorInterventionControlOutcomeFollow-upResults subspecialist vs. generalistOverall result
Studies comparing subspecialty care with general specialist care
 Baumer et al. [31]Diabetic specialist and caseload >40/yearNon-specialistMultiple mean of HbA1C1 year1.47 vs. 1.69 (P < 0.001)+
 Greenfield et al. [32]EndocrinologistPrimary care physician/internistDiabetic complications7 years−2 (−1.0, 0.6)=
Change in HbA1C1.0 (0.0–2.1)
 Greenfield et al. [33]Endocrinologist careInternistHbA1C <10%1 year1.32; CI: 0.67–2.59=
Blood pressure ≥ 140/901.06; CI: 0.71–1.59
LDL cholesterol0.69; CI: 0.46–1.04
 Levetan et al. [16]aEndocrinologistGeneralistReadmission rate3.5 years2 vs. 6% (P = ;0.03)+
 Schiel et al. [36]Centralized diabetic care by diabetologistsGeneral care by non-specialized physiciansDiabetic complications5 yearsNo effect=
 Zgibor et al. [39]Specialist careGeneralist careOvert nephropathy10 yearsRR: 0.4; CI: 0.2–0.9+/=
Proliferative retinopathyRR: 1.1; CI: 0.79–1.8
Lower extremity arterial diseaseRR: 0.95; CI: 0.4–1.1
Coronary artery diseaseRR: 0.65; CI: 0.4–1.1
Studies comparing (sub-) specialist care with primary care
 De Berardis et al. [23]Diabetic outpatient clinicPrimary careHbA1C >8%2 yearsOR: 0.92; CI: 0.61–1.32=
 Pellegrini et al. [35]Diabetic outpatient clinicPrimary careBlood pressure ≥160/901 yearRR: 1.92; CI: 1.10–3.35
 Ismail et al. [34]aSpecialist diabetic clinicPrimary care practice run buy GP or GP with special diabetic interestHbA1C3–6 yearsNo difference=
 Shah et al. [37]Specialist carePrimary careHbA1C?7.9 vs. 8.3 (P < 0.0001)+
 Sone et al. [38]Specialist outpatient clinicPrimary careHbA1C2 years7.2 vs. 8.4 (P<0.001)+
Mortality10 yearsHR: 0.31; CI: 0.10–0.98
 Zoppini et al. [40]Diabetic center care (periodic assessment of therapy and complications; educational courses)Usual care by family physiciansAll-cause mortality10 yearsRR: 0.87; CI: 0.81–0.94+
Other comparisons
 McAlister et al. [18]aOne or more consultation by endocrinologistExclusively seen by GPAll-cause mortality5 years1.17; CI: 1.08–1.27
 Uchigata et al. [17]One or more consultation of a diabetic centerNever consultationEnd-stage renal disease11–20 yearsHR: 0.19; CI: 0.05–0.78+
MortalityHR: 0.31; CI: 0.10–0.98
 Turchin et al. [19]High physician volume (volume used as a continuous variable in logistic regression model)Low volumeHbA1C <7%3 yearsOR: 0.96; CI: 0.93–1.0=
  • See explanations of symbols (+, − and ±) in Table 2 footnote. HbA1C, % of glycated hemoglobulin; CI, 95% confidence interval; RR, relative risk; OR, odds ratio; HR, hazard ratio; GP, general practitioner.

  • aDenotes a study of higher quality.

Compared with generalist care, subspecialist care resulted in better outcomes in two of the five included studies (Table 3). A better outcome (lower readmission rate) was also observed in the study of the highest quality. In this study, the readmission rate of patients treated by subspecialists was 2% in 3.5 years compared with 6% among those treated by a generalist [16]. Studies comparing specialist care with primary care revealed better outcomes for specialist care in three studies, better outcomes for primary care in one study and no difference in two studies (Table 3).

One study that examined the effect of consultation of a diabetic center observed lower 10-year mortality compared with those who did not consult the diabetic center (Table 3) [17]. In contrast, another study of higher quality examining the effect of consultation of an endocrinologist observed higher 5-year mortality compared with those exclusively seen in primary care [18]. The only study addressing the impact of physician volume on HbA1C outcomes did not detect an effect of volume [19].

Cystic fibrosis

We identified three articles that reported on the value of centralized treatment for cystic fibrosis. Two studies failed to adjust adequately for differences in case mix at baseline (Table 1) [20, 21]. In one of these, the cross-sectional study of Mahadeva et al. [21], the outcomes of cystic fibrosis patients treated both in a pediatric and adult cystic fibrosis center were compared with those who had not received specialized care at all. However, the authors did not report on the reasons why patients were treated in a cystic fibrosis center or not. Body mass index was better in patients treated both in a pediatric and adult specialized center (Table 4). In contrast, no effect was seen on forced expiratory volume (FEV1). Due to its cross-sectional design and the lack of case-mix adjustment, the study by Walters et al. [20] was also classified as weak. (Table 1). Finally, in the historical cohort study by Nielsen et al. [22], improvements in treatment over time might explain the superior results of centralized treatment in this study.

View this table:
Table 4

Results of studies evaluating the effect of treatment of cystic fibrosis in a high-volume or specialized center

AuthorInterventionControlOutcomeFollow-upResults center vs. non-center careOverall result
Mahadeva et al. [21]Pediatric center care and adult center care (n = 50)Neither pediatric CF care nor adult care in a CF center (n = 36)Body mass index5–7 yearsIncreased to 21.6 vs. 18.8 (P < 0.001)+/=
FEV153.5 vs. 50.7 (P = 0.8)
Nielsen et al. [22]CF centerStandard careSurvival20 yearsHR: 0.60; CI: 0.38–0.94+
Walters et al. [20]CF clinic exclusively for CF patientsGeneral clinicSymptom scorean.a.2.33 vs. 2.51 (P < 0.005)+
  • CF, cystic fibrosis; n.a., not applicable; CI, 95% confidence interval; FEV1, forced expiratory volume in a second. ‘+’, Centralized treatment better; ‘−’, usual care better; ‘+/=’, conflicting results or no effect.

  • aSummary score calculated as the average of five symptoms (breathlessness, cough, sputum, abdominal discomfort, fatigue).


In this systematic review of studies comparing the outcome of specialist treatment with the outcome of generalist treatment, we identified few studies of sufficient quality to detect meaningful differences in outcomes. Comparative studies addressing the effect of some sort of specialization (care by a subspecialist or in a specialized center) did not provide consistent results on the added value of treatment of rheumatoid arthritis, diabetes mellitus or cystic fibrosis in a high-volume or specialized center. Moreover, our review did not show a difference in the body of evidence for specialized treatment of cystic fibrosis as opposed to the body of evidence concerning rheumatoid arthritis and diabetes mellitus.

Substantial heterogeneity was observed between the included studies in process of care as well as in the outcomes studied. Some studies addressed the role of subspecialist care, while others addressed the value of treatment in a specialized center or the value of treatment by a multidisciplinary team. The comparison group varied between primary care in some studies and general internist care in other studies. Only one study among patients with diabetes mellitus addressed the value of subspecialists treating a high volume of patients.

It is difficult to draw firm conclusions from this heterogeneous set of articles. Nonetheless, pooling studies addressing the value of subspecialist care and those addressing the value of treatment in a specialized center did not result in more consistent results. Neither did the inconsistency decrease if we restricted our sample to higher quality studies. Another review examining the value of treatment in high-volume centers mainly focussed on surgically treated patients. Only one of the 22 diseases studied was a chronic disease (AIDS). A significant volume–outcome relationship was found between hospital volume and in-hospital mortality among patients with AIDS [5]. It is unknown, however, whether this finding could be generalized to other chronic diseases.

Our results might have been different if we had included studies reporting process measures as indicators of quality. Two of the included studies on diabetes reported higher quality for centralized treatment in terms of process measures, but no effect on glycolized hemoglobin [23] or even an adverse effect on survival [18]. In contrast, Turchin et al. [19] reported decreased adherence to guidelines for physicians caring for a high volume of patients in comparison with those with a lower caseload and no difference in outcomes.

Few included studies were RCT, the preferred study design when the effect of interventions is evaluated [9]. Moreover, the majority of the included observational studies had several flaws in study design, such as no adequate control of the results for baseline differences in potentially confounding factors (case mix). Many studies adjusted the results for differences in case mix that appeared some years after starting the treatment. This approach does not prevent and may even introduce bias. Even if results are corrected for case mix in observational studies, residual confounding may be present [11]. Patients treated by a specialist might have similar or even worse outcomes, because specialists generally treat the more severe cases. As a consequence, these shortcomings could have masked possible superior results of specialized treatment. On the other hand, given the unpredictability of the direction of this type of bias, the reported superior results in some studies might be spurious.

The studies included in this review varied largely in quality. If we would restrict our analyses to studies of sufficient quality, we would not be able to draw any conclusion. Moreover, focussing on studies of higher quality would not have changed our main conclusions. Although we judged the quality of many studies as weak, it is possible that this inferior quality merely reflects deficits in reporting of the study. As emphasized in the STROBE statement for observational studies [24], explicit reporting of key characteristics of study design is essential to be able to properly judge the quality of a study. As a consequence, our quality ratings might have looked different if all studies had reported all items in an appropriate way. It is, however, unlikely that this would have caused a systematic bias in one direction, leaving our main findings largely unchanged. Similarly, our review was also hampered by the limited information on the outcomes measured in some studies. We considered contacting the authors of those studies not a fruitful approach, because most included studies were rather old.

Another limitation is that the comparison group in the included studies largely differs. The quality of usual and standard care varies between countries and regions. If usual care is of high quality, it would be difficult to show added value of specialized treatment. Therefore, studies showing positive effects of centralization might also underline the lack of sufficient quality of usual or standard care.

A final issue to be raised is the likelihood of publication bias. We searched the literature using various keywords (MeSH terms) and free text. We could have missed some relevant articles, as we identified a number of articles only through the reference list of other articles and did not attempt to identify ‘grey literature’. Articles reporting null results or negative results usually have a higher chance to be missed [25]. However, because we identified several articles reporting inconsistent or negative results, publication bias does not seem to play an important role in our study.


The lack of an association between treatment of chronic diseases in a high-volume or specialized center and the outcome of medical care does not exclude the potential value of specialized centers. However, specific processes of care could be responsible for its added value rather than centralization of medical care per se [8]. This hypothesis is supported by a study on the impact of high-volume providers on the outcome of elderly patients with myocardial infarction. The differential use of proven effective medical treatment explained up to one-third of the survival benefit attributed to high-volume hospitals [26]. Similarly, increased use of endovascular treatment in high-volume hospitals appeared to explain a significant proportion of the volume–outcome relationship for abdominal aortic aneurysm surgery [27]. Rather than promoting treatment at high-volume centers, Halm et al. [8] recommended research directed at uncovering differences in processes of care between high- and low-volume providers in the Institute of Medicine report on the volume–outcome relationship. How outcome differences between groups can be explained and how improvements can be achieved has been illustrated in intensive care [28] and surgical services [29]. This approach may also reduce differences in outcome rates between specialized centers, as was observed after starting a prospective cystic fibrosis registry in North America [30]. Nonetheless, treatment of some chronic diseases in specialized centers might be justified to ensure that sufficient expertise is available and to improve cost efficiency. However, research directed at unraveling specific processes of care that explain differences in outcomes might be a more powerful tool in improving healthcare quality.


Financial support was received from the Dutch Ministry of Health, Welfare and Sport.


We thank Rikie Deurenberg for her assistance in searching the literature.


View Abstract