Participation identification and flow
Ambulance service researchers experienced difficulties in identifying study participants, as contemporaneous frequent caller lists were not available at all sites. While it was possible to identify eligible patients from call logs, it is not known how many of those patients were identified at the time (and therefore, how many of those patients in intervention areas were referred to case management, or received it), or if any eligible patients were missed.
Lower than anticipated case numbers meant three of the four ambulance services could not identify the anticipated number of eligible patients in case management areas after including patients already on the frequent callers list at the start of 2018. One ambulance service identified considerably more patients than anticipated in its intervention area. Ethical permissions did not allow data for all these patients to be extracted, nor was it feasible to do so. Instead, the list of eligible patients was sorted by postcode and date eligible for inclusion, then systematically sampled to ensure a wide geographic and temporal spread. A subsequent ‘top-up’ sample was selected the same way to compensate for the lower number of intervention patients identified in other sites. The remaining intervention cases were not included in analysis due to restrictions on data access.
Two ambulance services were unable to identify the number of patients expected in control sites, partially mitigated by extending these sites. Both extended sites were (individually) geographically contiguous. Another ambulance service identified more patients than anticipated in its control site; all of whom were included in the study.
Figure 2 shows the participant flow diagram for STRETCHED. Researchers at the participating ambulance services retrospectively identified a total of 1757 eligible patients (1103 intervention; 654 control) across 8 study sites (1 intervention and 1 control site per ambulance service).
FIGURE 2 STRETCHED patient flow. (Figure not reproduced here; see the original article.)
Data including patient identifiers, demographics, emergency ambulance service calls, ambulance dispatches, serious case reviews, and (where available) data on arrests and convictions were extracted for 1220 of these patients (569 intervention, 651 control) and linked with routine electronic data from NHS Digital and the SAIL databank.
Linkage rates to routinely collected data were extremely high, with no patients lost to due to failures in the linkage process. In total, 37 patients (19 intervention, 18 control) were excluded during data cleaning, an attrition rate of 3%. Most of these patients (33 in total; 15 in the intervention arm, 18 in the control arm) were found not to meet study inclusion criteria, while 4 intervention patients were excluded for data quality reasons. Further details are not available due to the risk of inadvertent disclosure through low counts in some subgroups. After these exclusions, the primary outcome analysis included data on a cohort of 1183 patients (550 in the intervention arm; 633 in the control arm).
Demographic and baseline data
Table 2 shows patient characteristics at baseline by study arm and ambulance service. Overall, the study cohort were relatively elderly, with a median age of 65 years (lower quartile 45, upper quartile 81), with slightly more females (51.9%) than males, and were predominantly white (81.7%). Just under two-thirds of patients (65.1%) lived in areas classified in the two most socioeconomically deprived quintiles, while only 5.2% lived in areas classified in the least socioeconomically deprived quintile.
| Patients identified | Intervention | Control | ||
|---|---|---|---|---|
| n = 550 | n = 633 | |||
| AS1 | 82 | (14.9%) | 72 | (11.4%) |
| AS2 | 77 | (14.0%) | 137 | (21.6%) |
| AS3 | 107 | (19.5%) | 248 | (39.2%) |
| AS4 | 284 | (51.6%) | 176 | (27.8%) |
| Age (years): median (lq – uq) | (n = 520) | (n = 583) | ||
| Overall | 60 | (41–80) | 69 | (49–82) |
| AS1 | 62 | (45–82) | 58 | (47–76) |
| AS2 | 50 | (38–71) | 63 | (47–82) |
| AS3 | 65 | (47–80) | 73 | (52–83) |
| AS4 | 60 | (39–81) | 69 | (48–81) |
| Gender: n (%) female | (n = 499) | (n = 572) | ||
| Overall | 266/499 | (53.3%) | 290/572 | (50.7%) |
| AS1 | 45/76 | (59.2%) | 30/60 | (50.0%) |
| AS2 | 26/59 | (44.1%) | 48/104 | (46.2%) |
| AS3 | 62/107 | (57.9%) | 123/248 | (49.6%) |
| AS4 | 133/257 | (51.8%) | 89/160 | (55.6%) |
| Ethnicity: n (%) non-whitea | 72/400 | (18.0%) | 75/412 | (18.2%) |
| Townsend quintilea,b | (n = 504) | (n = 577) | ||
| Q1 | 28 | (5.6%) | 28 | (4.9%) |
| Q2 | 61 | (12.1%) | 55 | (9.5%) |
| Q3 | 81 | (16.1%) | 126 | (21.8%) |
| Q4 | 99 | (19.6%) | 189 | (32.8%) |
| Q5 | 235 | (46.6%) | 179 | (31.0%) |
As noted above, there was considerable variation in numbers of patients classified by ambulance service and arm. This is particularly noticeable in the intervention arm, where over half the patients came from one site (AS4). Intervention site patients were also generally younger than control site counterparts in three out of the four participating ambulance services (AS2, AS3 and AS4). The proportion of female patients varied from 44.1% (AS2 intervention site) to 59.2% (AS1 intervention site) with no clear trend emerging across study arms or ambulance services. Data on ethnicity and deprivation quintile are not classified by ambulance service due to the risk of inadvertent disclosure from low counts in some subgroups.
There were more missing demographics data than anticipated. These data appear to be missing at random, but are not completely so – for instance, where demographics data were only sought from HES data sets, they were unavailable for those patients exercising the ‘opt-out’ option.
Baseline use
Table 3 shows emergency ambulance service use by patients for the 6 months before patients became eligible for inclusion in STRECHED. Call data are skewed towards lower numbers of calls with a small number of patients recording considerably more. Again, there was considerable variation in numbers of emergency ambulance service calls per patient classified by ambulance service and arm. The number of calls attended by ambulance and conveyed to hospital were both similar between arms, and with less variation across ambulance services.
| Intervention (n = 550) | Control (n = 633) | |||
|---|---|---|---|---|
| Emergency calls per patient: mean (SD) | ||||
| Overall | 11.61 | (17.13) | 15.48 | (88.93) |
| AS1 | 19.27 | (28.92) | 20.83 | (37.79) |
| AS2 | 19.03 | (28.73) | 35.53 | (187.86) |
| AS3 | 6.85 | (2.63) | 7.29 | (2.87) |
| AS4 | 9.17 | (7.71) | 9.23 | (9.24) |
| In-hours emergency calls per patient: mean (SD) | ||||
| Overall | 3.18 | (5.77) | 4.19 | (17.76) |
| AS1 | 4.59 | (8.03) | 6.00 | (13.31) |
| AS2 | 5.87 | (10.75) | 8.71 | (36.40) |
| AS3 | 2.10 | (1.86) | 2.32 | (1.81) |
| AS4 | 2.46 | (3.22) | 2.57 | (2.80) |
| Out-of-hours emergency calls per patient: mean (SD) | ||||
| Overall | 8.42 | (12.73) | 11.22 | (72.18) |
| AS1 | 14.68 | (22.88) | 14.83 | (25.29) |
| AS2 | 13.16 | (18.98) | 26.82 | (153.09) |
| AS3 | 4.75 | (2.44) | 4.97 | (2.89) |
| AS4 | 6.71 | (6.31) | 6.42 | (7.63) |
| Ambulance attendances per patient: mean (SD) | ||||
| Overall | 6.37 | (7.06) | 6.92 | (14.52) |
| AS1 | 7.60 | (7.27) | 7.07 | (7.55) |
| AS2 | 9.06 | (12.54) | 11.04 | (29.21) |
| AS3 | 4.15 | (2.18) | 4.52 | (2.68) |
| AS4 | 6.11 | (5.75) | 7.02 | (6.62) |
| Ambulance conveyances per patient: mean (SD) | ||||
| Overall | 3.14 | (4.39) | 2.97 | (3.67) |
| AS1 | 3.34 | (4.29) | 2.88 | (2.97) |
| AS2 | 4.88 | (7.21) | 4.06 | (6.20) |
| AS3 | 2.08 | (1.58) | 2.06 | (1.78) |
| AS4 | 3.01 | (3.99) | 3.47 | (2.88) |
Table 4 shows hospital use during the 6 months before each patient became eligible for inclusion in STRETCHED. Approximately two-thirds of patients recorded at least one emergency admission (753/1183; 63.7%), and/or outpatient appointment (804/1183; 68.0%). The proportion of patients recording emergency admissions, ED attendances, elective admissions and outpatient appointments were generally similar between arms; however, the overall mean number (per patient) of emergency admissions, elective admissions and outpatient appointments were all higher in intervention sites.
| Intervention | Control | |||
|---|---|---|---|---|
| Proportion of patients with at least one emergency admission; n (%) | ||||
| Overall | 351/550 | (63.8%) | 402/633 | (63.5%) |
| AS1 | 51/82 | (62.2%) | 33/72 | (45.8%) |
| AS2 | 36/77 | (46.8%) | 66/137 | (48.2%) |
| AS3 | 69/107 | (64.5%) | 179/248 | (72.2%) |
| AS4 | 195/284 | (68.7%) | 124/176 | (70.5%) |
| Number of emergency admissions per patient: mean (SD) | ||||
| Overall | 2.01 | (2.80) | 1.76 | (2.27) |
| AS1 | 2.07 | (2.84) | 1.79 | (3.27) |
| AS2 | 1.69 | (2.73) | 1.45 | (2.34) |
| AS3 | 1.42 | (1.65) | 1.98 | (2.21) |
| AS4 | 2.31 | (3.11) | 1.67 | (1.72) |
| Proportion of patients with at least one ED attendance: n (%) | ||||
| Overall | 458/550 | (83.3%) | 505/633 | (79.8%) |
| AS1 | 66/82 | (80.5%) | 52/72 | (72.2%) |
| AS2 | 62/77 | (80.5%) | 98/137 | (71.5%) |
| AS3 | 92/107 | (86.0%) | 191/248 | (77.0%) |
| AS4 | 238/284 | (83.8%) | 164/176 | (93.2%) |
| Number of ED attendances per patient: mean (SD) | ||||
| Overall | 5.10 | (7.38) | 4.40 | (6.92) |
| AS1 | 5.89 | (6.80) | 5.17 | (7.18) |
| AS2 | 8.31 | (9.29) | 6.09 | (11.29) |
| AS3 | 3.24 | (4.79) | 2.48 | (2.97) |
| AS4 | 4.69 | (7.50) | 5.49 | (5.62) |
| Proportion of patients with at least one elective hospital appointment: n (%) | ||||
| Overall | 90/550 | (16.4%) | 84/633 | (13.3%) |
| AS1 | 13/82 | (15.9%) | 10/72 | (13.9%) |
| AS2 | 12/77 | (15.6%) | 23/137 | (16.8%) |
| AS3 | 13/107 | (12.1%) | 32/248 | (12.9%) |
| AS4 | 52/284 | (18.3%) | 19/176 | (10.8%) |
| Number of elective hospital appointments per patient: mean (SD) | ||||
| Overall | 0.36 | (2.71) | 0.24 | (1.18) |
| AS1 | 0.95 | (6.74) | 0.22 | (0.61) |
| AS2 | 0.21 | (0.57) | 0.23 | (0.54) |
| AS3 | 0.22 | (0.81) | 0.24 | (1.02) |
| AS4 | 0.29 | (0.88) | 0.28 | (1.79) |
| Proportion of patients with at least one outpatient appointment | ||||
| Overall | 387/550 | (70.4%) | 417/633 | (65.9%) |
| AS1 | 59/82 | (72.0%) | 45/72 | (62.5%) |
| AS2 | 50/77 | (64.9%) | 92/137 | (67.2%) |
| AS3 | 61/107 | (57.0%) | 139/248 | (56.0%) |
| AS4 | 217/284 | (76.4%) | 141/176 | (80.1%) |
| Number of outpatient appointments per patient: mean (SD) | ||||
| Overall | 4.12 | (8.20) | 3.38 | (4.93) |
| AS1 | 7.06 | (16.70) | 2.71 | (3.46) |
| AS2 | 4.06 | (7.51) | 4.89 | (6.52) |
| AS3 | 1.80 | (2.61) | 1.77 | (2.53) |
| AS4 | 4.16 | (5.34) | 4.76 | (5.77) |
Comparison of outcomes
Primary outcomes
The primary outcome in STRETCHED is a composite indicator of mortality; emergency hospital admissions; ED attendances; and emergency ambulance service calls. Components were analysed individually and cumulatively by severity using logistic regression analysis adjusted for patient characteristics (ambulance service, age, sex, ethnic group, deprivation quintile) and service use (number of emergency ambulance service calls, emergency admissions, ED attendances, elective admissions, and outpatient appointments) for the 6 months prior to eligibility for inclusion in STRETCHED. The proportion of patients where an event was recorded and the mean number of events per patient during the 6-month follow-up are presented at an overall level in Table 5 (primary outcome components) and Table 6 (composite outcome). Outcomes for individual ambulance services are presented in Appendix 2.
| Intervention (n = 550) | Control (n = 633) | Adjusted comparisona | ||||
|---|---|---|---|---|---|---|
| Difference (95% CI) | p-value | |||||
| Mortalityb: n (%) | 58 | (10.5%) | 89 | (14.1%) | OR = 0.713 (0.465 to 1.093) | 0.121 |
| At least one emergency admission recorded: n (%) | 371 | (67.5%) | 422 | (66.7%) | OR = 1.114 (0.831 to 1.492) | 0.470 |
| Emergency admissions recorded: mean (SD) | 2.27 | (3.31) | 1.94 | (2.85) | Λ = 1.152 (0.984 to 1.349) | 0.083 |
| At least one ED attendance: n (%) | 423 | (76.9%) | 468 | (73.9%) | OR = 1.088 (0.763 to 1.551) | 0.642 |
| ED attendances recorded: mean (SD) | 5.96 | (10.67) | 5.13 | (10.34) | Λ = 0.973 (0.848 to 1.115) | 0.702 |
| At least one emergency ambulance service call: n (%) | 483 | (87.8%) | 545 | (86.1%) | OR = 1.197 (0.794 to 1.805) | 0.391 |
| Emergency ambulance service calls; mean (SD) | 17.17 | (28.65) | 15.35 | (26.43) | Λ = 1.062 (0.930 to 1.212) | 0.394 |
| Composite | Intervention (n = 550) | Control (n = 633) | Adjusted comparisona | |||
|---|---|---|---|---|---|---|
| Difference (95% CI) | p-value | |||||
| Mortality or emergency admission: n (%) | 384 | (69.8%) | 446 | (70.5%) | OR = 1.013 (0.748 to 1.372) | 0.933 |
| Mortality or emergency admission or ED attendance: n (%) | 464 | (84.4%) | 535 | (84.5%) | OR = 1.005 (0.675 to 1.495) | 0.981 |
| Mortality or emergency admission or ED attendance or emergency ambulance service call: n (%) | 526 | (95.6%) | 601 | (94.9%) | OR = 1.159 (0.595 to 2.255) | 0.665 |
We did not detect any statistically significant difference between arms for the overall composite outcome, or the individual components of death, emergency admissions, ED attendances or further emergency ambulance service calls.
The proportion of patients recording at least one component of the composite primary outcome was very high (95.3% overall). While this is partly attributable to 86.9% of patients making at least one further emergency ambulance service call, almost as many patients (84.4%) recorded an event in at least one other component of the composite measure.
The proportion of patients that died within the 6-month follow-up period was relatively high (10.5% intervention, 14.1% control, 12.4% overall). Although mortality did not differ significantly with study arm, it was strongly associated with age (p < 0.001), which does differ between arms. The most frequently recorded reasons for death, consistent with the age of the study population, are reported in Table 7.
| Cause | Intervention (n = 58 deaths) | Control (n = 89 deaths) | Comparison | ||
|---|---|---|---|---|---|
| Diseases of the circulatory system: n (%) | 12/58 | 20.7% | 30/89 | 33.7% | χ2 = 6.735, degrees of freedom = 3, p = 0.081 |
| Malignant neoplasms | 19/58 | 32.8% | 15/89 | 16.9% | |
| Diseases of the respiratory system | 12/58 | 20.7% | 15/89 | 16.9% | |
| Other | 15/58 | 25.9% | 29/89 | 32.6% | |
Secondary outcomes
Table 8 presents secondary study outcomes, including the proportion of patients referred for case management; the proportion of patients that no longer met the frequent caller threshold at any point (and the proportion of those patients that subsequently met the threshold), the proportion and mean number of outpatient appointments, elective appointments, the mean number of ambulance attendances and conveyances, and evidence of harm (defined as at least one adverse event, arrest, or conviction recorded, a complaint made by/on behalf of the patient, or a request for a serious case review). Outcomes were assessed for the 6 months after each patient was first eligible for inclusion in STRETCHED.
| Intervention (n = 550) | Control (n = 633) | Adjusted comparisona | ||||
|---|---|---|---|---|---|---|
| Difference (95% CI) | p-value | |||||
| Referred for case managementb: n (%) | 39/97 | (40.2%) | N/A | N/A | N/A | N/A |
| Patients declassifiedc: n (%) | 538/550 | (97.8%) | 626/633 | (98.9%) | OR = 0.619 (0.228 to 1.681) | 0.347 |
| Declassified patients subsequently meeting the frequent caller criteria: n (%) | 179/538 | (33.3%) | 183/626 | (29.2%) | OR = 1.075 (0.800 to 1.446) | 0.631 |
| Patients with at least one elective appointment: n (%) | 81/550 | (14.7%) | 54/633 | (8.5%) | OR = 1.802(1.198, 2.711) | 0.005 |
| Number of elective appointments: mean (SD) | 0.30 | (1.65) | 0.12 | (0.48) | Λ = 1.720 (1.229 to 2.407) | 0.002 |
| Patients with at least one outpatient appointment: n (%) | 390/550 | (70.9%) | 428/633 | (67.6%) | OR = 1.056 (0.767 to 1.455) | 0.737 |
| Number of outpatient appointments: mean (SD) | 4.91 | (9.28) | 3.61 | (5.43) | Λ = 1.046 (0.905 to 1.210) | 0.548 |
| Ambulance attendances: mean (SD) | 9.24 | (13.90) | 7.75 | (11.96) | Λ = 1.050 (0.918 to 1.200) | 0.493 |
| Ambulance conveyances: mean (SD) | 4.18 | (7.49) | 3.43 | (5.39) | Λ = 1.055 (0.915 to 1.217) | 0.469 |
Where data were available, approximately two-fifths of eligible patients (40.2%) were referred for case management, of whom 69.2% are known to receive it. However, only one ambulance service was able to provide these data for all intervention area patients, and a second service provided data for a subset of patients only. The other two services were unable to provide these data due to lack of access to contemporaneous frequent caller lists, as previously noted. This may impact the generalisability of study findings.
Almost all patients (98.4%) no longer met the frequent caller threshold at some point during follow-up (i.e. there was at least one point where the patient had not made 5 or more calls in the previous month or 12 or more calls over the previous 3 months). However, this does not take into account, for example, any extended hospital stays that may impact call frequency. Approximately one-third of these patients (intervention 33.3%; control 29.2%) subsequently met the call threshold again during the follow-up period, with no statistically significant difference observed between study arms.
The mean number of ambulance attendances and conveyances were similar between arms. Just over half of emergency ambulance service calls resulted in attendance (53.8% intervention; 50.4% control); < 25% of calls resulted in conveyance to hospital (24.3% intervention; 22.3% control). The most common reasons for calling were ‘fall’ (6.5%), ‘sick person’ (5.2%) and acute coronary syndrome (4.7%).
Similar numbers of patients recorded at least one outpatient appointment during the follow-up period in both arms. More patients recorded at least one elective appointment in intervention areas compared with control areas (14.7% intervention; 8.5% control; p = 0.005); the mean number of elective appointments was similarly higher in intervention areas (0.30 events per person in intervention areas; 0.12 events per person in control areas; p = 0.002).
There is only sparse data available on harms (adverse events, complaints made to the ambulance service, serious case reviews, arrests and convictions) both prior to and during the follow-up period – participating ambulance services reported that data for arrests and convictions were generally not available. On grouping together extremely low numbers of these different events, there are a total 5 such events in the intervention arm, and 11 in the control arm – no formal statistical comparison was undertaken due to uncertainty over the completeness of the data.
Further discussion of these results, in the context of findings from other strands of the study, is presented in the discussion chapter of this report. As STRETCHED was not powered to detect differences between arms within individual ambulance services, summaries of outcomes by ambulance service are presented in Table 18 (primary outcome and components) and Table 19 (secondary outcomes) in Appendix 2, while Appendix 3 contains further details of data processing, multiple imputation methods and statistical modelling of the primary outcome.