Literature search
The electronic database search resulted in 13820 citations and an additional 16 citations from hand searching key systematic reviews. After exclusion of duplicate records, 6416 articles were screened by titles and abstracts with 201 articles determined to be eligible for full-text assessment. Of these, 85 studies met the eligibility criteria and were included in our systematic review. Flowchart of the selection process from initial identification to inclusion is shown in Fig. 2. Main reasons for exclusion included patients treated in non-primary care settings, not meeting minimum PCMH components or focused on intervention other than PCMH model, lack of control group, and other reasons (list of excluded articles; see Appendix 2).
Descriptive data synthesis
The characteristics of included studies are presented in Appendix 3–4. Of the 85 studies included in the review, 78 studies were RCTs [13, 14, 16, 18–20, 22, 36–106] and 7 studies were of non-RCTs, including quasi trials [17, 21, 107, 108] or cohort studies with a control group [109–111]. The 85 studies enrolled a total of 60617 patients with sample sizes ranging from 40 to 8366. Whilst 79 studies had sufficient data for quantitative data synthesis, 6 studies [81, 85, 95, 97, 103, 107] did not have usable data and therefore, the findings were narratively summarised.
The common inclusion criteria for all 85 studies was primary care patients with diagnosis of one or more chronic conditions, whereas the predominant reason for exclusion was patients with cognitive impairment and terminal illness. More than half the studies (52%) were conducted in the United States. The mean age of patients ranged between 30 and 83 years. In terms of gender distribution, most studies had a slight female predilection except for studies conducted in Veterans Affairs (VA) primary care settings [16, 50, 52, 53, 56]. The duration of follow-up varied from 3 to 48 months.
Quality assessment and risk of bias
Quality assessment and risk of bias for individual studies are reported in Appendix 5–8. The overall quality of studies ranged from ‘fair’ to ‘good’ by meeting at least 60% of items on the checklist. Two studies [62, 104] were rated as poor due to general lack of information on randomisation, unclear methodology, and clarity of results. Given the nature of PCMH-based intervention, most trials employed a cluster randomisation method where a group of patients were seen by the same GP or same general practice providing PCMH care. Thereby, blinding of patients or GPs was not applicable and, as a result, items related to blinding were not necessarily graded down. However, only 32 studies reported blinding of outcome assessment whilst other studies were graded down in quality. The quality of evidence across included studies assessed using GRADE approach is presented in Tables 1–2.
Table 1
GRADE assessment of randomised controlled trials reporting effectiveness of PCMH vs standard GP care on outcomes of interest
Outcome | No of studies | Risk of bias | Inconsistency | Indirectness | Imprecision | Publication bias | GRADE quality of evidenceþ |
Depression | 30 | Serious | Serious | Not serious | Not serious | Undetected | Moderate‡ |
Quality of Life | 19 | Serious | Not serious | Not serious | Not serious | Undetected | Moderate‡ |
Blood pressure | 12 | Serious | Not serious | Not serious | Not serious | Undetected | Moderate‡ |
Glycated Hemoglobin | 8 | Serious | Serious | Not serious | Not serious | Undetected | Low‡¶ |
LDL Cholesterol | 3 | Serious | Serious | Not serious | Not serious | Undetected | Low‡¶ |
Total Cholesterol | 1 | Serious | - | Not serious | Not serious | Undetected | Low‡^ |
Hospital admissions | 3 | Serious | Not serious | Not serious | Not serious | Undetected | Moderate‡ |
Self-management (PACIC scores) | 3 | Serious | Serious | Not serious | Not serious | Undetected | Low‡¶ |
Cost-effectiveness | 18 | Serious | Serious | Not serious | Not serious | Undetected | Low‡¶ |
þHigh quality: Further research is very unlikely to change our confidence in the estimate of effect; Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate; Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate; Very low quality: We are very uncertain about the estimate. |
‡Most studies did not blind participants or personnel as it was not practical. Therefore, we did not downgrade for these risks/uncertainties. However studies not reporting blinding of outcome assessment were downgraded in quality. |
§Most studies did not mention allocation concealment strategies. |
¶Significant level of heterogeneity within results (I2 between 80–90%) |
^Single study – Inconsistency not applicable |
Table 2
GRADE assessment of non-randomised studies reporting effectiveness of PCMH vs standard GP care on outcomes of interest
Outcome | No of studies | Risk of bias | Inconsistency | Indirectness | Imprecision | Publication bias | Certainty |
Depression | 1 | Serious | - | Not serious | Not serious | Undetected | Low†‡^ |
Quality of Life | 2 | Serious | Not serious | Not serious | Serious | Undetected | Moderate†‡^ |
Blood pressure | 1 | Serious | - | Not serious | Not serious | Undetected | Low†‡^ |
Glycated Hemoglobin | 1 | Serious | - | Not serious | Not serious | Undetected | Low†‡^ |
LDL Cholesterol | 1 | Serious | - | Not serious | Serious | Undetected | Low†‡^ |
HDL Cholesterol | 1 | Serious | - | Not serious | Not serious | Undetected | Low†‡^ |
Total Cholesterol | 1 | Serious | - | Not serious | Not serious | Undetected | Low†‡^ |
Hospital admissions | 2 | Serious | Not serious | Not serious | Not serious | Undetected | Moderate†‡^ |
Cost-effectiveness | 1 | Serious | - | Not serious | Not serious | Undetected | Low†‡^ |
†Because of the nature of the quasi-experimental designs risk of bias is unavoidable |
‡Most did not blind participants or personnel as it was not practical. Therefore, we did not downgrade for these risks/uncertainties. However studies not reporting blinding of outcome assessment were downgraded in quality. |
^Single study – Inconsistency not applicable |
Depression outcomes
Meta-analysis of thirty-one studies [13, 14, 18, 19, 36, 38, 40, 42, 43, 46, 50, 51, 53, 55, 57, 63, 67, 68, 70, 76, 78, 83, 84, 86–88, 91, 93, 100, 102, 109] of patients with minor or major depression episodes after PCMH-based care reported significant improvement in depression scores compared to patients with standard primary care. With the exceptions of three studies [46, 91, 102], twenty-two studies reporting changes in mean differences (continuous data) of depression scores showed significant reduction with a pooled SMD of -0.24 (95% CI -0.35, -0.14; p-value < 0.001) (Fig. 3).
Six studies reported that PCMH care was associated with significantly increased odds of remission of depression with pooled OR 1.79 (95% CI 1.46, 2.21; p-value < 0.001) (Fig. 3). Additionally, one other study [85] reported significant improvements among patients with anxiety and mood disorders with an effect size of 0.30 (95% CI 0.05, 0.55; p-value = 0.02) compared to standard care. Given most studies consistently reported improvements, the GRADE of evidence was classified as moderate quality (Tables 1 and 2).
Quality of life outcomes
Twenty-two studies [18, 19, 21, 22, 41, 46, 49–51, 53, 59, 68, 72, 76, 86, 89, 91, 100, 102, 105, 106, 108] evaluated the effectiveness of PCMH-based care on HRQoL (overall, physical component and mental component). Patients enrolled in PMCH-based care reported small but significant improvements in HRQoL compared to standard care with a pooled SMD of 0.10 (95% CI 0.04, 0.15; p-value < 0.001) (Fig. 4). Additionally, one other study [85] reported significant improvements with an effect size of 0.38 (95% CI 0.13, 0.63; p-value = 0.003). Moderate heterogeneity was observed among included studies (I2 = 57%), but test for sub-group differences were not significant. The GRADE of evidence was classified as moderate quality (Tables 1 and 2).
Blood pressure outcomes
Thirteen studies [16, 17, 39, 42, 45, 61, 64, 68, 71, 82, 90, 94, 96] reported on the effect of PCMH care on blood pressure outcomes. Six studies reported that PCMH care was associated with significantly increased odds of BP control with pooled OR 2.03 (95% CI 1.56, 2.65; p-value < 0.001) (Fig. 5). Seven studies reported significant improvements in systolic blood pressure (SBP), in favour of PCMH care, with pooled estimates of SMD − 0.15 (95% CI -0.29, -0.01; p-value = 0.03). Similar reduction was observed across five studies reporting on diastolic blood pressure (DBP), but the pooled estimate of SMD − 0.12 (95% CI -0.27, 0.02; p-value = 0.09) failed to meet significance (Fig. 5). The GRADE of evidence was classified as moderate quality (Tables 1 and 2).
Glycated haemoglobin outcomes
Ten studies [16, 17, 39, 43, 64, 68, 71, 77, 82, 96] reported on the effect of PCMH care on HbA1c outcomes. Three studies reported that PCMH care was associated with increased odds of glycaemic control with pooled OR 2.37 (95% CI 0.86, 6.51; p-value = 0.100). However, the pooled estimate was not statistically significant (Fig. 6). The substantial heterogeneity of 87% in the three studies reporting ORs was due to a shorter follow-up duration of 3 months reported by Bogner et al [43] compared to the other two studies which had follow-up duration of 12 to 13 months. Seven studies reported significant improvements in HbA1c, in favour of PCMH care with pooled estimates of SMD − 0.26 (95% CI -0.43, -0.08; p-value = 0.004) (Fig. 6). Given the substantial amount of heterogeneity, the GRADE of evidence was classified as low quality (Tables 1 and 2).
Cholesterol outcomes
For LDL-cholesterol outcomes, five studies [17, 64, 68, 71, 96] reported significant improvements in favour of PCMH care with pooled SMD of -0.16 (95% CI -0.33, -0.00; p-value = 0.05) compared to standard GP care. Test for subgroup difference between follow-up and change scores showed no statistical significance (I2 = 16.8%, p-value = 0.27) (Fig. 7A). For total cholesterol outcomes, two studies [17, 82] reported a non-significant increase in total cholesterol with a pooled SMD of 0.07 (95% CI -0.08, 0.23; p-value = 0.34) (Fig. 7B). The GRADE of evidence of both LDL and total cholesterol outcomes were classified as low quality given the limited number of studies (Tables 1 and 2).
Hospital admissions
Five studies [20, 21, 48, 54, 111] reported that PCMH care was associated with significant reduction in hospital admissions compared to standard care with pooled OR 0.83 (95% CI 0.70, 0.98; p-value = 0.02) (Fig. 8). Additionally, one study [110] reported a reduction in mean hospital admission rates related to diabetic complications 12 months after PCMH based care compared to standard care. Nonetheless, the change in mean difference failed to meet statistical significance. The GRADE of evidence was classified as moderate quality (Tables 1 and 2).
Self-management outcomes
Three studies [14, 72, 89] reported significant improvements in self-management scores in favour of PCMH care compared to standard care with pooled estimates of SMD 0.24 (95% CI 0.03, 0.44; p-value < 0.001) (Fig. 9). Given the substantial amount of heterogeneity (I2 = 83%), the GRADE of evidence was classified as low quality (Tables 1 and 2).
Economic outcomes
A total of 18 studies [13, 22, 37, 44, 46, 52, 58–60, 65, 66, 69, 73, 79, 80, 92, 98, 108] reported cost-effectiveness of PCMH-based models of care compared to standard care. To avoid bias in analysis, all currencies were converted to US Dollars at the time of the respective trials and cost effectiveness was measured in terms of incremental cost of intervention. The incremental cost of PCMH care was small but significantly higher than standard care with a pooled estimate of 0.17 (95% CI 0.08, 0.26; p-value < 0.001) (Fig. 10). The substantial heterogeneity of 81% was due to higher costs of intervention reported by Bosanquet et al [46]. The GRADE of evidence was classified as low quality (Tables 1 and 2).
A summary of results from meta-analyses (where possible) and individual studies from randomised and non-randomised controlled trials are presented in Table 3.
Table 3
Summary of findings from meta-analyses (where possible) or individual studies from randomised and non-randomised controlled trials
Outcome | No of studies | No of participants | Effect size (95% CI) | p-value | Q statistic | I2 | Egger’s test p-value‡ | Citations | Figure |
Randomised controlled trials |
Depression | 24 6 | 7255 1520 | SMD − 0.24 (-0.35, -0.14) OR 1.79 (1.46, 2.21) | < 0.001 < 0.001 | 78.3 3.58 | 76% 0% | 0.275 0.608 | [13, 14, 18, 19, 36, 38, 40, 42, 43, 46, 50, 51, 53, 55, 57, 63, 67, 68, 70, 76, 78, 83, 84, 86–88, 91, 93, 100, 102, 109] | 3 |
Quality of Life | 22 | 12370 | SMD 0.12 (0.09, 0.15) | < 0.001 | 57.38 | 51% | 0.556 | [18, 19, 21, 22, 41, 46, 49–51, 53, 59, 68, 72, 76, 86, 89, 91, 100, 102, 105, 106, 108] | 4 |
Blood pressure BP control Systolic BP Diastolic BP | 6 6 5 | 1202 1947 1836 | OR 2.03 (1.56, 2.65) SMD − 0.08 (-0.17, 0.01) SMD − 0.12 (-0.27, 0.02) | < 0.001 0.09 0.10 | 5.30 8.97 7.82 | 6% 44% 49% | 0.347 0.737 0.260 | [16, 39, 42, 45, 61, 64, 68, 71, 82, 90, 94, 96] | 5 |
Glycated haemoglobin Glycaemic control HbA1c | 3 6 | 726 2044 | 2.37 (0.86, 6.51) SMD − 0.21 (-0.30, -0.12) | 0.001 < 0.001 | 15.00 27.75 | 87% 82% | NA 0.405 | [16, 39, 43, 64, 68, 71, 77, 82, 96] | 6 |
LDL Cholesterol | 4 | 1086 | SMD − 0.25 (-0.37, -0.13) | < 0.001 | 1.64 | 0% | NA | [64, 68, 71, 96] | 7A |
Total Cholesterol | 1 | 888 | SMD 0.00 (-0.13, 0.13) | 1.00 | NA | NA | NA | [82] | 7B |
Hospital admissions | 3 | 4770 | OR 0.90 (0.80, 1.03) | 0.12 | 0.67 | 0% | NA | [20, 48, 54] | 8 |
Self-management (PACIC scores) | 3 | 2440 | SMD 0.24 (0.03, 0.44) | 0.02 | 11.48 | 83% | NA | [14, 72, 89] | 9 |
Cost-effectiveness | 17 | 12612 | SMD 0.17 (0.07, 0.26) | 0.001 | 87.84 | 82% | 0.206 | [13, 22, 37, 44, 46, 52, 58–60, 65, 66, 69, 73, 79, 80, 92, 98] | 10 |
Non-randomised trials |
Depression | 1 | 314 | SMD − 0.22 (-0.45, 0.01) | 0.06 | NA | NA | NA | [109] | 3 |
Quality of Life | 2 | 833 | SMD − 0.08 (-0.21, 0.06) | 0.28 | 0.94 | 0% | NA | [22, 108] | 4 |
Blood pressure Systolic BP | 1 | 727 | SMD − 0.30 (-0.45, -0.16) | < 0.001 | NA | NA | NA | [17] | 5 |
Glycated haemoglobin | 1 | 727 | SMD − 0.20 (-0.35, -0.06) | 0.006 | NA | NA | NA | [17] | 6 |
LDL Cholesterol | 1 | 727 | SMD 0.06 (-0.09, 0.20) | 0.43 | NA | NA | NA | [17] | 7 |
HDL Cholesterol | 1 | 727 | SMD 0.15 (0.00, 0.29) | 0.05 | NA | NA | NA | [17] | - |
Total Cholesterol | 1 | 727 | SMD 0.16 (0.01, 0.30) | 0.04 | NA | NA | NA | [17] | 8 |
Hospital admissions | 2 | 912 | OR 0.63 (0.48, 0.83) | 0.001 | 0.02 | 0% | NA | [21, 111] | 9 |
Cost-effectiveness | 1 | 358 | SMD 0.19 (-0.01, 0.40) | 0.07 | NA | NA | NA | [108] | 10 |
NA – not applicable |
‡Egger’s test was conducted only for outcomes with at least 6 studies. |
Publication bias
Six or more articles with similar outcomes were inspected for publication bias visually by using funnel plots and statistically by determining the significance from Egger’s test of asymmetry. Visual inspection of included studies reporting similar outcomes did not indicate any obvious sign of asymmetry (Figs. 11 and 12). Consistent with visual findings, no evidence of publication bias was detected with Egger’s test, as all outcomes had p > 0.05, showing evidence of funnel plot symmetry (Table 3).