Table 3. Results of Included Studies.

Study ID	Results	Notes
Rich 1993	(All reported as intervention gp. vs. control gp.) Primary endpoints: Proportion readmitted within 90 days: 21/63 (33.3%, 95% CI 21.7, 44.9) vs. 16/35 (45.7%, 95% CI 29.2, 62.2) NS (Not clear if this is unplanned re-admissions or all re-admissions). Total number of days hospitalised within 90 days: mean 4.3 days per patient (SE 1.1, 95% CI 2.1, 6.5) vs. 5.7 (SE 2.0, 95% CI 1.8, 9.6) NS, P value not given. Also looked at: Days to first re-admission: mean 31.8 days (SE 5.1, 95% CI 21.8, 41.8) vs. 42.1 (SE 7.3, 95% CI 27.8, 56.4) NS, P value not given. Proportion readmitted within 90 days in the moderate risk subgroup: 11/40 (27.5%, 95% CI 18.7, 41.3) vs. 10/21 (47.6%, 95% CI 26.2, 69.0) NS P = 0.1, Fisher’s exact test Proportion readmitted within 90 days in the high risk subgroup: 10/23 (43.5% 95% CI 23.2, 63.8) vs. 6/14 (42.9%, 95% CI 17.0, 68.8) NS, P value not given. Total number of days hospitalised within 90 days in the moderate risk subgroup: 3.2 (SE 1.2, 95% CI 0.8, 5.6) vs. 6.7 (SE 3.2, 95% CI 0.4 to 13.0) NS, P value not given. Total number of days hospitalised within 90 days in the high risk subgroup: 6.3 (SE 2.3, 95% CI 2.0 to 10.6) vs. 4.2 (SE 1.4, 95% CI 1.5 to 6.9) NS, P value not given. Days to first re-admission in the moderate risk subgroup: 35.1 (SE 9.0, 95% CI 17.5 to 52.7) vs. 28.6 (SE 7.2, 95% CI 14.5, 42.7) NS, P value not given. Days to first re-admission in the high risk subgroup: 27. 8 (SE 3.5, 95% CI 20.9 to 34.7) vs. 60.2 ( SE 10.5, 95% CI 39.6, 80.8) P < 0.05 in favour of usual care, Mann-Whitney rank sum test	Comment on statistical analyses: Since it was a feasibility study this study was not powered to show differences in its primary endpoints. Overall 38% (95% CI 28 to 47) of patients were readmitted within 90 days. No information on survival supplied, all patients were followed up for 90 days - presumably all survived
Rich 1995	(All reported as intervention gp. vs. control gp.) Primary endpoint Survival for 90 days without re-admission: 91 (64.1%) vs. 75 (53.6%), absolute difference 10.5% (95% CI −0. 9, +21.9) P = 0.09, NS Secondary endpoints Total number of readmissions in 90 days (all cause): 53 vs. 94, P = 0.02, Wilcoxon rank-sum test Patients with at least one re-admission in 90 days (all cause): 41 (28.9%) vs. 59 (42.1%), absolute difference 13.2% (95% CI 2.1, 24.3), P = 0.03 Patients with more than one re-admission in 90 days (all cause): 9 (6.3%) vs. 23 (16.4%), absolute difference 10.1% (95% CI 2.8, 17.4), P = 0.01 Total number of days in hospital during follow up: 556 vs. 865 Mean days in hospital during follow up per patient: 3. 9 (SD 10.0) vs. 6.2 (SD 1.4), P = 0.04, Wilcoxon rank-sum test Total number of heart failure re-admissions in 90 days: 24 vs. 54, P = 0.04, Wilcoxon rank-sum test QOL (CHFQ) at baseline and 3 months in subset of 126 patients (67 intervention patients, 59 control): Total score mean change between baseline and 90 days +22.1 ( SD 20.8) vs. +11.3 (SD 16.4) P 0.001 (QOL also increased significantly more in the intervention group compared to the control group in each of the 4 subscales of the CHFQ) Costs of care for study patients: $4,815 mean costs per patient vs. $5,275 Also reported Deaths in follow up period: 13 (12.1%) vs. 17 (9.2%), NS Admission to long term care facility during follow up: 5 vs. 6 Survival for 90 days without re-admission in those surviving initial hospitalisation: 66.9% vs. 54.3 6%, absolute difference 12.6% (95% CI 1.1, 24.1) P = 0.04 Re-admission rates in 9 months after follow up (all cause): NS difference Heart failure re-admission rates in 9 months after follow up: 57 vs. 80, P 0.08, NS Substudy of medication compliance: Average percentage of pills taken correctly for each medication at 30 days after discharge (method 1): 87.9% (SD 12.0, range 33.3% to 100%) vs. 81.1 (SD 17.2, range 23.1% to 100%), P 0.003 Total number of pills taken correctly divided by total number of pills that should have been taken (method 2): 87.5% (SD 12.6, range 35.1 to 100%) vs. 80.9 (Sd 16.7, range 23% to 100%) Overall mean compliance rate for sub study population: method 1, 84.6% (SD 15.1, range 23.1% to 100%); method 2, 84.3% (SD 15, range 23% to 100%), P = 0. 004	Comment on statistical analyses: The study had adequate power - with 282 patients it was the largest heart failure specific study identified in this review and statistical review suggests that all statistical analyses were appropriate. Not clear how heart failure re-admissions and non CHF re-admissions were distinguished. Not clear how similar two groups are at baseline in subset of patients with QOL data. Not clear whether the cost data was just on 57 patients and how those patients were selected and how similar the two groups were at baseline
Cline 1998	(All reported as intervention gp. vs. control gp.) “Main outcomes” (primary endpoint not specified): Mean days to re-admission in survivors at one year: 141 (87) vs. 106 (101), P < 0.05 (see comment below). Mean days in hospital in survivors at one year: 4.2 (7. 8) vs. 8.2 (14.7) NS, P= 0.07 (unequal variance). Health care costs in survivors at one year: (mean costs) 2294 US$ vs. 3594 US$, NS, P= 0.07. Also looked at: Deaths at one year: 24 (31%) vs. 31 (28%), NS, (test statistic not given) Deaths at 90 days (Cline 2001): 8/80 vs. 17/110, P given as <0.001, erroneous, our estimation P = 0.3 Death or at least one re-admission to hospital (composite end point) at 12 months: 56 (70%) patients vs. 79 (72%) patients, NS, (test statistic not given) Death or at least one re-admission to hospital (composite end point) at 90 days: At three months 53 (66%) vs. 61(56%) NS Number of patients surviving to one year who were readmitted: 22 (39%) vs. 43 (54%) NS, P =0.08 Mean no. of hospitalisations per patients surviving to one year: 0.7 (SD 1.1) v 1.1 (SD 1.8) NS , P =0.08 Outpatient visits (not clear how defined): 3.6 (3.2) vs. 4.0 (3.4) NS Treatment at one year, % on ACEI: 41 (75%) vs. 41 (52%) P <0.05 Treatment at one year, % on all other HF drugs: all differences NS The quality of life in heart failure questionnaire, Nottingham health profile and patients’ global self assessment: all differences NS at one year, (test statistics not given)	Comment on statistical analyses: No sample size calculation given, but our post hoc calculations suggest that this study had adequate power. We note an apparent error on page 444, the text says that says 56/79 patients died or were readmitted, Table 2 says 56/79 patients survived some with readmission. Elsewhere the text suggests 46/74 died or were readmitted. The mean time to re-admission in patients who survived to one year was longer in the intervention than in the control group (mean days to re-admission 141 (SD 87) vs. 106 (100) is given as having P <0.05, but this result is contradicted by the non-significant result of the robust log-rank test which tests the outcome time to death or readmission. Outcome data on re-admissions, days in hospital and costs all on survivors at one year not on whole group. Health care costs involved many assumptions. Study had some before and after analyses that are not reported here
Ekman 1998	(All reported as intervention gp. vs. control gp., where appropriate.) Main endpoints: Proportion of patients aged >65 years who were eligible for the study: 15% (95% CI 13, 17) NB this included those who refused. Proportion of patients in the intervention group who did not visit the HF nurses: 23 (29%, 95% CI 19-39%) (11 died, 11 fatigued/unwilling, 1 institutionalised) Functional class: change in NYHA classification between baseline and 6 month follow up int. gp. - cont. gp. = 0.08 (95% CI −0.19, 0.35), NS (test statistic not given) Hospitalisations: mean difference intervention gp. -control. gp. = −0.1 (95% CI −0.5 to 0.3), NS (test statistic not given) Deaths: 19 (24%) vs. 15 (19%), NS (test statistic not given) Hospital days: intervention gp. - control. gp. difference in mean days 8 (SD 26), difference in median days 4, P= 0.29 (NS) Mann-Whitney U test Also reported: Survival at 6 months without re-admission: 25 (32%) vs. 31 (39%), NS (test statistic not given) 7 (9%) intervention patients made spontaneous visits to HF nurses as a result of clinical deterioration and were admitted to hospital. 4 (median) phone contacts with nurses per intervention patient. 22 (28%) intervention patients made spontaneous phone contacts to HF nurses. 18 carers/relatives (23%) of intervention patients made spontaneous phone contacts to HF nurses	Comment on statistical analyses: Since there were only 79 patients in each arm of this feasibility study it may have lacked the statistical power to detect differences in its main outcomes. No sample size calculation was given (feasibility study).
Rainville 1999	(All reported as intervention group. vs. control group.) Primary endpoint: Number of patients who died (all causes) or were read-mitted with HF at one year: 5 vs. 14, P < 0.01 probably Chi squared test (NB See statistical comment below.) Time to re-admission for HF or patient death: Significantly longer in intervention group, P < 0.01 log rank test. Also reported: Number of patients readmitted with HF at one year : 4 vs. 10, P 0.05 probably Chi squared test (NB See statistical comment below.) Deaths (all cause) at one year: 1 v 4 (test statistic not given) Total no. of re-admissions at one year: 20 v 26 NS, (test statistic not given) Change in functional health assessment score (Dartmouth COOP charts): no significant change at 30 or 90 days for either group (no test results given)	Comment on statistical analyses: Very small sample size. Inappropriate statistical tests used with small sample sizes. (Chi-square test should not be used here, the correct test is Fisher’s exact test.)
Stewart 1999a	(All reported as intervention gp. vs. control gp.) Primary endpoint: Frequency of unplanned re-admissions plus all cause out-of-hospital deaths during 6 months follow up: 77 primary events vs. 129 primary events, event rates per month 0.20 (95% CI 0.14-0.26) vs. 0.40 (0.24-0.56), P = 0.02 (test not clear) (NB see statistical comment). Other endpoints: Frequency of unplanned re-admissions alone at 6 months: 68 vs. 118, event rates per month 0.14 (95% CI 0.10-0.18) vs. 0.34 (0.19-0.49), P = 0.03 (NB see statistical comment). Out of hospital deaths at 6 months: 9 v. 11, NS All cause deaths at 6 months: 18 vs. 28, P = 0.098 Number of patients remaining event free (i.e. death or re-admission) at 6 months: 51 vs. 38, P = 0.04 Total unplanned days in hospital at 6 months: 460 v. 1174, event rates per month 0.9 (0.6-1.2) vs. 2.9 (1.93.9), P = 0.01(NB see statistical comment). Total hospital based costs (including inpatient, out patient and emergency services): median per patient per month, A$ 252 [IQR 37-1179] vs. 438 [42-2172], P = 0.16. Cost of hospital and community based health care, (sample of patients only): median per patient per month, A$ 400 vs. 380, P = 0.91. Change in Minnesota living with heart failure questionnaire between baseline and 3 & 6 months (random sample of 68 patients): intervention group significantly bigger fall in score than control group at 3 months (higher scores indicate impaired QOL but clinical significance of change seen not clear), no significant difference in scores of survivors at 6 months. Change in Australian version SF-36 between baseline and 3 & 6 months (random sample of 68 patients): no differences seen in mental health scores, change in physical health scores at 3 months significantly higher in intervention group (clinical significance not clear) but no difference in survivors at 6 months. Also looked at: Difference in probability of survival at 18 months: P = 0.1 Frequency of unplanned re-admissions alone at ‘the end of follow up’ (around 18 months): 118 vs. 156, event rates per month 0.15 (0.11-0.19) vs. 0.37 (0.19-0.55), P = 0.053 Total elective days in hospital at 6 months: 87 vs. 25, P = 0.13 Total unplanned days in hospital at ‘the end of follow up’ (around 18 months): 875 vs. 1476, event rates per month 1.1 (0.8-1.4) vs. 2.7 (1.6-3.7), P = 0.04 Regression analysis showed that assignment to intervention group was a borderline, independent predictor of survival, P = 0.046, Cox proportional hazards model	Comment on statistical analyses: A rationale for the sample size was provided and the sample size appears to have been adequate. The primary end-point is unusual and does not correspond to any well known statistical test. The statistical tests used to analyse multiple events are unclear and it is not certain which test they used to analyse their primary endpoint. Other points: Not clear how re-admission was determined to be unplanned or planned. Frequency distribution of unplanned re-admissions in the two groups suggests that the difference in unplanned admissions was predominantly amongst those relatively few patients who had three or more admissions in the 6 month follow up period - most of these were in the control group. Intervention group patients accumulated more elective days in hospital (87 vs. 25 P = 0.13) the majority for surgical procedures delayed whilst patient clinically unstable. However, similar proportions of unplanned re-admissions associated with a primary diagnosis of heart failure in each group: 34 (50%) intervention vs. 58 (49%) controls. 88/100 intervention patients received intervention 2 died and 10 withdrew after initial consent. After initial home visit immediate review by primary care physician or cardiologist was requested for 33 patients, 42 intervention patients had a flexible diuretic regimen introduced, 19 patients had greater pharmacy contact arranged and 23 patients had new or increased home support services arranged
Jaarsma 2000	(All reported as intervention gp. vs. control gp.) Please note: The authors of this study make adjustments to their findings both for attrition and multiple testing. For clarity mean scores, SDs and P values for the different scales are not shown in this table unless the authors have stated by that the findings are significant. Measures of QOL Heart Failure Functional Status Inventory: difference NS at 3 months and at 9 months Symptom occurrence: difference NS at 1, 3 and 9 months Change in symptom severity and distress from baseline: considerable attrition at both 3 and 9 months (e.g. for symptom severity only 26/58 responses in intervention group, 42/74 in control group at 9 months), differences NS at 3 months and NS at 9 months after attempt to adjust for attrition by attributing change score of zero to missing values. Psychosocial Adjustment to Illness Scale: differences NS at 3 and 9 months. Cantril’s Ladder of Life: Patients often stated that they had difficulty with this scale resulting in several missing values, differences NS at 1 month (results for 3 and 9 months not given because patients had such difficulty using the scale*). Measures of self-agency and self care behaviour Self care abilities, Appraisal of Self-Care Agency Scale: differences NS at baseline, 3 and 9 months follow up. Self care behaviours: difference NS at baseline and 9 months follow up. At 1 month: 13.8 (SD 3.4) vs. 12. 2 (2.9) P < 0.001, at 3 months: 11.6 (SD 3.1) vs. 10.2 (3.3) P < 0.005. Health care resource use Hospital readmissions (on all patients i.e. 84 intervention and 95 control patients), average days in hospital per patient at 9 months follow up: 9 vs. 9, NS, P value and test not given. Patients with at least one hospital re-admission at 9 months follow up: 31 (37%) vs. 47 (50%), P = 0.06, Chi squared test Cardiac readmissions at 9 months follow up, mean days per patient: 5.1 days (SD 11) vs. 7.1 days (SD 15), NS, P value and test not given Patients with at least one cardiac hospital re-admission at 9 months follow up: 24 (29%) vs. 37 (39%), P = 0. 1, Chi squared test Patients with at least one cardiac hospital re-admission at 9 months follow up: 24 (29%) vs. 37 (39%), P = 0. 1, Chi squared test Patients with at least one hospital re-admission at 1 month follow up: no significant differences between the two groups Hospital readmissions, average days in hospital per patient at 3 months follow up: 5.1 vs. 5.1, NS, P value and test not given. Patients with at least one hospital re-admission at 3 months follow up: 22 (26%) vs. 29 (31%), NS, P value and test not given. Cardiac readmissions at 3 months follow up, mean days per patient: 18 (21 %) vs. 23 (24%) NS, P value and test not given. Hospital readmissions, average days in hospital for cardiac readmission per patient at 3 months follow up: 3. 0 vs. 4.1, NS, P value and test not given. Also reported: Deaths during 9 months follow up : 22 vs. 16 NS	Comment on statistical analyses: Rationale for sample size given. The exact statistical tests used in the final analysis were unclear. Other points: 186 patients were enrolled in the study, 7 died before discharge from the index admission, by 9 months 47 (26%) of the remaining 179 had died or dropped out, the data for those who remained had a large number of missing values. Differences in self care behaviour scores significantly better at 1 and 3 months in intervention group but mean differences very small and clinical significance unclear
Blue 2001	(All reported as intervention gp. vs. control gp.) Primary endpoint at around 12 months: Number of patients with death from all causes or hospital admission for heart failure: 31 vs. 43, hazard ratio = 0.61 (95% CI 0.38 to 0.96), P = 0.03 Secondary endpoints at around 12 months: Death: 25 v. 25 NS Number of patients with death from all causes or all cause hospital re-admission: 52 vs. 61, hazard ratio = 0. 72 (95% CI 0.49 to 1.40) NS, P = 0.075 Number of patients with hospital re-admission (all causes): 47 vs. 49, P = 0.27, NS Number of patients with hospital re-admission for worsening HF: 12 vs. 26, P = 0.004 Also looked at (at around 12 months): Number of admissions per patient per month (all causes): 0.124 vs. 0.174, hazard ratio = 0.71 (95% CI 0.54 to 0.94) P = 0.018 Number of admissions per patient per month (worsening HF): 0.027 vs. 0.069, hazard ratio = 0.40 (95% CI 0.23 to 0.71) P = 0.0004 Mean days spent in hospital (all causes): 10.3 (SD 19. 0) vs. 16.7 (24.1), P = 0.08 Mean days spent in hospital (worsening HF): 3.43 (12. 2) vs. 7.46 (16.6), P = 0.005	Comment on statistical analyses: The size of the study was only just adequate for statistical power based on a calculation for 12 months follow up. Some of the statistical results are presented in an ambiguous way. The study might have benefited from a further exploration of the data with some sensitivity analyses. For example, the primary endpoint includes deaths that occurred in hospital after randomisation. As it happens there were more deaths in the control group than in the intervention group (6 vs. 1), if the analysis is re-done excluding these in pre-discharge hospital deaths the primary endpoint is no longer significant
Capomolla 2002	(All reported as intervention group vs. control group.) Primary outcomes: Relative risk of cardiac death or urgent heart transplantation: RR0.17, favouring intervention, (95% CI 0.06, 0.66) Deaths from cardiac causes: 3/112 (3%) vs. 21/122 (17%), P 0.0007 Total number of hospital readmissions at mean 12 (SD3) months follow up: 13 vs. 78, P<0.00001 (NB not clear if these readmissions are because of haemodynamic instability as stated earlier in the paper.) Total number of patients with at least one rehospitalisation during follow up: 8 vs. 35 P <0.05 Secondary outcomes: QOL (time trade off method): 0.72 (SD 0.17) vs. 0. 63 (SD 0.22) P < 0.008. (ie intervention patients were willing to trade 10 years of their present health for 7.2 years of excellent health, whereas control patients were willing to trade 6.3 years of their present health. (NB only change within the two groups reported.) NYHA functional class: only reported as before and after findings and error in table showing the results of NYHA functional class at one year see below. Also looked at: Mean total care management costs (ie group mean read-mission costs and day hospital costs): $167,785 vs. $178,553, no standard deviations or tests given. Cost utility ratio of the two strategies: $2,244 vs. $2, 409 Incremental analyses showed a cost saving of$1,068 for each QALY gained with the intervention. Urgent heart transplant during follow up: 1 vs. 0.	Comment on statistical analyses: No sample size provided but study appears to have adequate power and the statistical tests employed are appropriate. There is a serious error in Table 3 on page 1263: NYHA classifications are given on 112/112 usual care patients at one year and on 113/112 day hospital patients at one year despite the fact that we are told that cardiac death occurred in 21/122 patients in the community group and 3 patients in the day hospital group. Comment on cost utility analysis: The study is not from a societal perspective as stated since it does not consider all the costs to the patient (i. e. travel, time off work, home help). The information for the cost utility analysis is poorly presented. ‘Total costs’ should be presented with ‘average’ costs since from the averages it is hard to cross check calculations. It would also have been helpful to see the break down of the costs and quantities used, so the source of these ‘totals’ could be known. A cross check of their calculations suggests that they are consistent but the total costs figures do not seem to be consistent with the averages reported. It would also be helpful to know how they calculated the total QALYs per programme to check to what extent they have allowed for the timing of the deaths within the year. Other points: Not stated how deaths from cardiac causes were identified. Not stated how readmissions because of haemodynamic instability were identified. Total number of deaths in each group not given. Not clear if QOL was same at baseline for both groups. Not all the 112 patients in the intervention received all the components of the intervention: 76% received education and physical training; 47% received cardiovascular risk stratification; 45% received tailored therapy; 19% received multidisciplinary intervention. There were 49 ‘open access interventions’ in the intervention group, these included interventions which would have required admission in the control group
Doughty 2002	(All reported as intervention gp. vs. control gp.) Primary endpoints: Event free survival, time to first hospital admissions or all cause death: P = 0.33 (NS), Kaplan-Meier HRQL, Change in MLHFQ total score from baseline to 12 months between the two groups: P = 0.1 (NS) (change in physical score −11.1 vs. −5.8, P = 0.015, change in emotional score −3.3 both groups, P= 0.97, NS) Secondary endpoints: All cause hospital re-admission rates at 12 months: 1.37 re-admissions per patient per year vs. 1.84 (method of calculation not given), rate difference = 0.47 per patient per year (95% CI 0.16, 0.78) . All cause total hospital bed days at 12 months: 1074 vs. 1170 NS, test statistic not given, 12.3 bed days per patient per year vs. 13.9, mean difference in bed days per patient = 1.6 (95% CI 0.51, 2.7) (method not given) Re-admissions for heart failure at 12 months: 36 vs. 65 NS Also reported: All cause deaths at one year: 19 (19%) vs. 24 (25%) Medication: trend (P = 0.052) for intervention group to be on higher ACEI dose at 12 months, no other significant differences Mean time to 1st hospital re-admission: 102 (SD 104) vs. 122 (SD 116) P =0.4, NS. (method not given) Total all cause re-admissions at 12 months: 120 vs. 154 NS, P value not given (method not given) First all cause re-admissions at 12 months: 64 vs. 59 NS, P value not given (method not given) Subsequent all cause re-admissions 56 vs. 95, P = 0.015 (Fishers exact test - test inappropriate) All cause hospital bed days first re-admissions: 546 vs. 444 (no statistical test result given) All cause bed days during subsequent re-admissions: 528 vs. 726 P = 0.0001 (test method not given) First heart failure related re-admissions at 12 months: 21 vs. 23 NS, test statistic not given Subsequent heart failure related re-admissions at 12 months: 15 vs. 42 P <0.05 (Fishers exact test - test inappropriate) Total hospital bed days for heart failure related re-admissions at 12 months: 358 vs. 561 NS, test statistic not given Hospital bed days for first heart failure related re-admission at 12 months: 219 vs. 195 NS, test statistic not given Hospital bed days for subsequent heart failure related re-admission at 12 months: 139 vs. 366 P = 0.0001 (test method not given)	Comment on statistical analyses: Trial terminated early apparently because inadequate power to detect difference in primary endpoint. An accompanying editorial article states: “The study was actually prematurely stopped. A provisional estimate of 180 patients per group was made but the final sample size was calculated after 100 patients had been followed for 6 months. The event rate was found to be higher than expected but there was no difference between the two groups for the combined primary endpoint of death or re-admission. Projection of the observed effect size suggested that an order of magnitude of more patients would have been required to achieve a result reaching statistical significance, but even this would probably have had little clinical significance. The follow-up of patients already recruited was completed to allow data for total admissions and quality of life to be analysed” (Cunningham 2001). The explanation of the statistical analysis for the analysis of multiple admissions lacks clarity. Some of the data is continuous but since admissions and bed days are likely to be highly skewed the t test, whose use is mentioned in the paper, would be inappropriate. Also the use of Fisher’s exact test to compare subsequent readmissions does not seem appropriate since this test cannot be used to analyse multiple events. Other comments: Baseline values of MLHFQ not given. 60% of the intervention group attended the first group educational session, 40% attended the six month educational session
Harrison 2002	(All reported as intervention group vs. control group.) Primary outcome on 157 patients: MLHFQ total score at baseline, mean (SD): 44.8 (18. 5) vs. 44.6 (19.5), P = 0.9 (NS) MLHFQ total score at 6 weeks, mean (SD): 27.3 (19. 1) vs. 37.5 (20.3), P for difference = 0.002 MLHFQ total score at 12 weeks, mean (SD): 25.8 (19. 4) vs. 38.4 (18.2), P for difference < 0.001 (Also significant improvements in physical dimension at 6 and 12 weeks and in emotional dimension at 6 but not 12 weeks.) Proportion of patients at 6 weeks with at minimally significant difference in total MLHFQ score (5 points) from baseline, worse, same, better: intervention group 3 (4%), 18 (23%), 58 (73%), control group 17 (22%) , 17 (22%), 42 (55%), P = 0.002. Proportion of patients at 12 weeks with at minimally significant difference in total MLHFQ score from baseline worse, same, better: intervention group 6 (8%), 7 (9%), 65 (83%) v. control group (29%), 10 (13%), 44 (58%), P = 0.001 Secondary outcomes: QOL measured with SF-36: SF 36 total scores not given. Data difficult to interpret. Proportion of patients making an emergency room visit (all causes) during the 12 weeks after discharge (on the 157 patients who were followed up): 29% vs. 46%, P= 0.03, Chi squared test. Proportion of patients re-admitted to hospital (all causes) during the 12 weeks after discharge (on the 157 patients who were followed up): 23% vs. 31%, P = 0. 26, NS	Comment on statistical analyses: Sample size calculation given, sample size appears satisfactory for primary endpoint. Authors mention lack of power to detect a difference in readmissions. Statistical analyses appear appropriate.
Kasper 2002	(All reported as intervention gp. vs. control gp.) Primary endpoint: Total number of CHF hospital admissions plus all cause deaths: 50 in 102 patients vs. 72 in 98 patients, P = 0. 09 (NS) log transformed t test, P = 0.03 Poisson model comparison (see comment below). Secondary endpoints: Deaths at 6 months: 7 in 102 patients vs. 13 in 98 patients P = 0.14 (NS) log-rank test Re-admissions for CHF: 43 admissions in 26 patients vs. 59 admissions in 35 patients, P = 0.09 (NS) log transformed t test, P = 0.03 Poisson model comparison. All cause hospital admissions plus all cause deaths: 84 in 102 patients vs. 109 in 98 patients, P = 0.13 (NS) log transformed t test, P = 0.04 Poisson model comparison. Event free survival (death or re-admission at 6 months) : P for difference = 0.12 (NS) QOL: change in Minnesota Living with Heart Failure Q (MLHFQ) total score at 6/12 from baseline, mean, median −28.3, −28 vs. −15.7, −15, P = 0.001 Wilcoxon test (lower MLHFQ score = better) Functional status: change in Duke Activity Status Score at 6/12 from baseline, 1.1, 1 (mean, median) vs. 0.8, 1 v, P = 0.44 (NS) Wilcoxon test. (Duke Activity Status Index also NS). Process measures at 6 months: Proportion of patients with systolic dysfunction receiving target vasodilators: 74/80 vs. 43/7, P < 0.001 Dietary compliance “good”or “average”: 65/94 vs. 38/ 85, P = 0.002 At goal weight: 47/94 vs. 17/85, P = 0.001. Medication compliance: NS difference. Cost data: Mean costs: $16,182 vs. $8,789 (NS) 75th centile $6,527 vs. 10,898 (NS)	Comment on statistical analyses: Poisson model analyses suggested by the Oversight, Data and Safety Monitoring Committee before patient enrolment. However the Poisson model does not hold for these data. This is because the results are over dispersed. This means that there are a few patients with higher numbers of readmissions than would be expected using a Poisson model, so if the data are analysed using Poisson model then an incorrectly small SP results. If the results are analysed using a Poisson model where the over dispersion is accounted for then the SP is 0.11, very near the value from the log-rank test of 0.13. Other comments: Cost difference remained NS if three high cost patients all randomised to the intervention group (two transplant patients and one patient who died shortly after the study) were excluded from cost comparison
Krumholz 2002	(All reported as intervention gp. vs. control gp.) Primary outcome: Death or all cause readmission at one year : 25 vs. 36, RR 0.69, 95% CI 0.52, 0.92, P = 0.01 Risk of HF or other CVD readmission, or death (intervention vs. control): HR0.51, 95% CI 0.29, 0.90, P=0. 02) (Cox proportional hazards model adjusted for age; sex; history of HF and admission serum creatinine) Secondary outcomes: Deaths: 9/44 (20%) vs. 13/44 (30%) RR0.69 (95% CI 0.33 to 1.45), P=0.33, NS. Total readmissions in one year: 49 vs. 80, P=0.06, test not given. Total HF readmissions: 22 vs. 42, P=0.07, NS, test not given. Multiple readmissions 12/44 (27%) vs. 21/44 (48%) RR 0.57 (95% CI 0.33 to 0.99) P=0.05. Risk of HF readmission or death (intervention vs. control): HR 0.52, 95% CI 0.28, 0.98, P=0.04 (Cox proportional hazards model adjusted for age; sex; history of HF and admission serum creatinine). Number of patients experiencing HF or other CVD readmission or death: 22/44 (50%) vs. 35/44 (80%). RR 0.63, 95%CI 0.46, 0.86, P=0.004. Number of patients with at least one heart failure read-mission or death: 18/44 (41%) vs. 30/44 (68%) RR 0. 6 95% CI 0.41,0.89, P=0.01. All cause hospital days readmitted, mean (SD): 10.2 (S 16.8) vs. 15.2 (SD 17.5), P=0.09 test not given. HF or other CVD hospital readmission days, mean (SD): 6.3 (SD 9.2) vs. 12.3 (SD 14.3), P=0.03 test not given. HF hospital days readmitted, mean (SD): 4.1 (SD 6.4) vs. 7.6 (12.1), P=0.1 NS, test not given. Costs: All cause readmission costs, mean per patient: $14,420 vs. $21,935, P=0.02, test not given HF or other CVD readmission costs, mean per patient: $8,888 vs. $18,421, P=0.01, test not given HF readmission costs, mean per patient: $5,232 vs. $9, 575, P=0.04, test not given	Comment on statistical analyses: The sample size was small and no power calculation was supplied. The statistical analysis seems appropriate, although it is not clear which tests were used for multiple admissions. Cost of care package $530.
McDonald 2002	(All reported as intervention gp. vs. control gp.) Primary endpoint: Number of patients with death or re-admission for HF within 12 weeks 4 v. 12, P = 0.04 (Fisher’s exact test) . 95% CI for Odds ratio 0.07-0.84, not clear how OR was generated Also reported: Deaths at 3 months 3 vs. 3, NS Number of patients with a re-admission for HF within 12 weeks 1 v. 9, statistical test not supplied Number of patients with re-admission for HF within 12 weeks 1 v. 11, P = < 0.01 95% CI for odds ratio 0. 01-0.53, not clear how OR was generated Number of patients with death or re-admission for HF within one month (from McDonald 2001) 0/35 v.O/35, NS Clinical condition (NYHA class), LVEF, BP, U&E: all NS difference QOL (not stated how measured): NS Process measures at 3 months: Patient knowledge of HF 16.3 (SD 2.7) vs. 13.1 (SD 2. 2) (presume mean scores out of 20), P = < 0.01 Patient knowledge of diet 8.3 (SD 2.1) vs. 6.6 (SD 1. 9) (presume mean scores out of 10), P = < 0.01 Carer knowledge of HF: NS difference Mean doses of frusemide, digoxin or ACEI: NS difference	Comment on statistical analyses: No sample size calculation was performed. Other comments: Data on all cause admissions not supplied. There are discrepancies in two tables between the interim paper (McDonald 2001) and the final results paper (McDonald 2002): the number of patients who died in their index admission or were excluded because of co-morbidity compromising their survival went down between the two papers, as did baseline demographics for the number of patients with an idiopathic aetiology for their HF and with previous admissions for HF within one or three months of the index admission. In addition the rate of recruitment appears to be very different in these two reports
Riegel 2002	(All reported as intervention gp. vs. control gp.) Primary endpoint: (See statistical analyses comment) Mean number of hospitalisations with HF per patient at 3 months: 0.17 (SD 0.43) vs. 0.31 (SD 0.64), P=0. 03, analysis of covariance. Mean number of hospitalisations with HF per patient at 6 months: 0.21 (SD 0.5) vs. 0.41 (SD 0.77), P=0.02, analysis of covariance. Other endpoints: (See statistical analyses comment ) Mean number of all cause hospitalisations per patient at 3 months: 0.45 (SD 0.73) vs. 0.61 (SD 0.88) P=0. 25, analysis of covariance. Mean number of all cause hospitalisations per patient at 6 months: 0.62 (SD 0.88) vs. 0.87 (SD 1.1) P=0.11, analysis of covariance. Proportion of patients re-admitted with HF at 3 months: 14.6% vs. 22.8%, P=0.06, multiple logistic regression Proportion of patients re-admitted with HF at 6 months: 17.7% vs. 27.6%, P=0.06, multiple logistic regression Proportion of patients readmitted with all causes at 3 months: 33.8% vs. 41.2%, P=0.40, multiple logistic regression Proportion of patients readmitted with all causes at 6 months: 43.1% vs. 50.0%, P=0.49, multiple logistic regression Mean number of heart failure related hospital bed days at 3 months: 0.9 (2.3) vs. 1.6(3.9), P=0.56 multiple linear regression Mean number of heart failure related hospital bed days at 6 months: 1.2 (3.1) vs. 2.1 (4.6), P = 0.05, multiple linear regression Mean number of all cause hospital bed days at 3 months: 2.6 (5.0) vs. 3.5 (7.2), P = 0.35 multiple linear regression Mean number of all cause hospital bed days at 6 months: 3.5 (6.6) vs. 4.8 (8.3), P = 0.23 multiple linear regression Mean time to rehospitalisation at 6 months: 128.5 (SD 68.6) vs. 115.7 (SD 68.6), P=0.32, Proportion of patients with multiple readmissions at 6 months: 13.1%. vs. 22.8%, P=0.07 Mean emergency department visits at 6 months: 0.14 (SD 0.45) vs. 0.11 (SD 0.94), P=0.58 Patient satisfaction at 6 months: NB available on only 184 of the 242 patients the researchers attempted to survey, number in intervention and control groups not given, 22.88 (SD 2.85) vs. 21.66 (SD 3.44), P=0.01, clinical importance of1.22 point difference in this scale unclear, scale not validated. Also looked at: Cost of acute care at 6 months: $1192(SD 3674) vs. $2186 (SD 6729), P=0.07 Proportion of patients alive at 6 months: 87.7%. vs. 86%, P= not stated	Comment on statistical analyses: For the primary outcomes and other hospitalisation data the statistical tests used appear to be are inappropriate as the data are not normally distributed. Another major criticism of this paper is that it gives no information about the randomisation clusters: how many there were; how many physicians in each; how many patients in each on average; and how many clusters were finally analysed. This information is necessary to assess generalisability, bias, robustness of analysis and importance of adjusting for clustering. A related point is that if the physicians were matched then if one of a pair refused to participate, the other should also have been omitted and it is not clear that this has been done. The sample size calculation appears to take no account of clustering, which will result in an underestimate of numbers needed. However, the number of patients estimated is about the same as the number of physicians, so it is possible that cluster sizes were very small in which case the sample size calculation would not be much altered. (It is not possible to tell how big cluster sizes are because we are not told how many clusters were lost to follow-up.) Cost of intervention was $443.
Laramee 2003	(All reported as intervention gp. vs. control gp.) Primary endpoint: (results on 131 intervention and 125 control patients) Number of patients with a readmission (all causes) during 90 days after discharge: 49 (37%) vs. 46 (37%), P >0.99, NS Secondary endpoints: Adherence to the treatment plan: intervention group adhered to treatment plan better than usual care group at both 4 and 12 weeks with regard to daily weights, checking for oedema, low salt diet and fluid intake recommendations (however differences were small). Patient satisfaction: Mean scores on patient satisfaction survey were significantly higher in intervention group (however differences were small). Medications at 12 weeks (results on 128 intervention and 113 control patients): Taking ACEIs or ARBs: 108 (84%) vs. 90 (80%), P = 0.40, NS Taking beta blocker: 89 (70%) vs. 70 (62%), P = 0.22, NS Taking target does of ACEI of ARB: 64 (63%) vs. 42 (49%), P = 0.08, NS Target does of beta blocker: 27 (32%) vs. 18 (29%), P = 0.72, NS Also looked at: (results on 131 intervention and 125 control patients) Number of patients with a CHF readmission during 90 days after discharge: 18 (14%) vs. 21 (17%), P = 0.49, NS Number of patients with a cardiac readmission during 90 days after discharge: 15 (11%) vs. 10 (8%), P = 0. 40, NS Number of patients with more than one readmission: 7% vs. 8%, P = 0.83. Mean (SD) LOS in hospital for patients with at least one readmission: 6.9 (6.5) vs. 9.5 (9.8) P = 0.15, NS Median (interquartile range) LOS in hospital for patients with at least one readmission: 5 (2-8) vs. 7 (2-10) P = 0.37, NS. Deaths: (results on 141 intervention and 146 control patients) 13 (9%) vs. 15 (10%), P = 0.84, NS Cost data: (results on 135 intervention and 127 control patients who completed the 12 week study period) Total costs means medians: $23,054 vs. 25,536 P = 0. 39 NS; $15.979 vs. $18,662 P = 0.14 NS Total readmission costs means: $5,253 vs. $5,163 P = 0.96 NS Total outpatient costs means: $ 1,552 vs. C $1,307 P = 0.28 Initial admission costs means: $ 16,119 vs. $19,081 P = 0.18	Comment on statistical analyses: No sample size calculation given. Statistical analyses appear appropriate. Other comments: Readmission and LOS outcomes not given on ‘patients whose participation was terminated early and were not known to have been readmitted’. Outcome assessment not masked, possibility of bias in patient satisfaction scores
	gp = group patient, U&E = urea and electrolyte levels, * information from personal communication with author