Abstract
Background Most hospital admissions are for older patients, who are more likely to be medically complex. Primary care risks after hospital discharge and readmission trajectories are relatively unexplored.
Aim To estimate the frequency and nature of errors and harms linked to discharge summary processing in general practice for patients aged ≥65 years; and to explore associations with readmission and healthcare utilisation, using a novel primary care data source.
Design and setting Retrospective cohort study using electronic health records (EHRs) from seven purposively sampled general practices in the West Midlands, UK.
Method EHRs of patients aged ≥65 years discharged between October 2022 and October 2023 were reviewed. Outcomes included post-discharge healthcare utilisation and costs, frequency and type of discharge actions, error rates (failures to complete requested actions), harms (severity, attribution, and preventability), and 90-day readmissions. Multivariable logistic regression identified predictors of errors, harms, and readmission.
Results Within the cohort of 263 discharged patients, 186 (70.7%) accessed post-discharge care; with 47 out of the 263 patients (17.5%; 95% confidence interval = 13.7 to 23.0) having related readmissions within 90 days. In total, 551 actions were requested in 160 discharge summaries. For those patients who required an action, 20.6% experienced error and 4.4% experienced harm. Most harms resulted in hospital readmission and three out of eight were preventable in primary care. Error and harm disproportionately affected patients who had dementia or a recorded carer.
Conclusion General practice should review their processes for responding to patient discharge information in order to improve patient safety post-discharge. Further research into tools to assist practices with transitions is warranted.
How this fits in
Older patients frequently use general practice post-discharge care at point of need. Older patients (particularly those with carers or a dementia diagnosis) are at risk of harm from post-discharge primary care. Error rates in processing hospital discharge summaries in general practice appear to be improving in the last decade but harm remains. Recognition of a funded general practice post-discharge care episode and research into interventions for this cohort are crucial given the emphasis on the shift from secondary to primary care and the UK ageing population.
Introduction
Transition from hospital to home is a recognised point of vulnerability for older adults with multimorbidity, frailty, and polypharmacy.1,2 In England, people aged ≥65 years account for >6 million hospital admissions annually, including 2.6 million emergency admissions.3 While national guidance emphasises safe discharge and coordinated follow-up,4,5 discharge transitions remain a persistent risk,6 repeatedly featuring in patient concerns.7 In this article, the authors revisit the epidemiology of discharge-related harm in primary care in a 10-year update to their prior work.1
Discharge summaries are frequently a source of communication failure between hospital and general practice: often delayed, unclear, or missing key clinical details making safe and timely care difficult.8–10 The authors’ prior work found that nearly half of discharged older patients’ summaries requiring general practice action were not fully processed, and 8% were linked to moderate or severe patient harm.1 Patients and carers also report feeling unprepared for discharge and unsure who to contact once home, leaving general practice to manage unresolved issues.7
Though hospital-to-home transitions occur within a wider system of community and out-of-hospital services, general practice provides the ongoing, longitudinal follow-up for most older adults after discharge. Despite this, most research has focused on hospital-led (‘transitional care’) interventions to reduce readmissions.11 Far less is known about how post-discharge care is organised within general practice; as a result, few interventions exist.12 Recent ethnographic work describes wide variation in discharge summary handling, reflecting local workflows and staffing rather than shared protocols.13 Increasing frailty and ‘left shift’14 of activity from hospital to primary care place an unknown burden on general practice in the post-discharge space.
Medication reconciliation is a particular risk, with 15% of older adults experiencing medication-related harm within 8 weeks of discharge, much of it preventable and often leading to further healthcare use.15 Interventions that reduce readmissions16 are not routinely integrated into general practice.
This study aimed to:
describe patterns of (primary and urgent) care utilisation following discharge in patients aged ≥65 years, including economic analysis;
explore the relationship between primary care follow-up and hospital readmission, including the feasibility of a novel use of primary care electronic health records (EHRs) data to examine post-discharge outcomes; and
identify general practice-originated errors and harms associated with discharge.
This study provides a 10-year update to the authors’ previous retrospective records analysis of discharge-related harm in general practice.1
Method
Study design
A retrospective cohort study using anonymised EHRs from patients registered at seven general practices in the West Midlands, UK. Practices were purposively selected from those expressing interest to provide maximum variation in list size, rural/urban setting, ethnic diversity, and socioeconomic profile (based on practice-level Index of Multiple Deprivation [IMD] data) (Table 1).
Table 1. Study practice characteristics Record inclusion criteria
Records from all patients aged ≥65 years discharged from hospital from October 2022 to October 2023 were included. Patients with <90 days of EHR data post-discharge were excluded.
Data collection
The third and last authors, an academic pharmacist and an academic GP, respectively, reviewed EHRs onsite at each practice. The first six post-discharge contacts in general practice were extracted in chronological order. Data quality was ensured by working as a dyad with live discussion between the clinician reviewers to prevent misclassification of failures. Eligible records were identified using local search tools, anonymised at the point of extraction, and entered into a structured Microsoft Excel case report form (Box 1). Methods were aligned with those of the authors’ previous study1 to allow comparison.
Box 1. Variables extracted from electronic health records, by domain Data analysis
Descriptive statistics summarised general practice and patient characteristics, post-discharge care use, and discharge actions. Frequencies and proportions were reported for categorical variables; means and standard deviations (SDs) (or medians and interquartile ranges [IQRs]) for continuous data. Multivariable logistic regression identified predictors of: failure to complete discharge summary actions (errors); harms resulting from these errors; and readmission within 90 days. Variables were selected for clinical relevance and compared to the authors’ 2014 data.1 Odds ratios (ORs), 95% confidence intervals (CIs), and P-values were reported. The likelihood ratio test assessed overall significance of categorical predictors. Subgroup analyses examined outcomes by age group and admission type. Analyses were conducted using R statistical software. Healthcare service use was costed descriptively using standard NHS reference sources.17–19 Mean and total costs were reported by service type and patient subgroup. Harm severity was graded by the clinical academic reviewers using established World Health Organization scales,17 preventability was graded according to previously used scales.1,18,19
Results
General practice and patient characteristics
In total, seven general practices participated, with list sizes ranging from 5700 to 26 500 patients (Table 1). Patients aged ≥65 years comprised 5.6%–30.0% of practice populations. Practices varied in socioeconomic deprivation, ethnicity, rurality, and Care Quality Commission (CQC) rating. The ratio of patients to whole-time equivalent GPs ranged from 1:883 to 1:2470.
The cohort comprised 263 patients with a mean age of 77.4 years; 140 were male (53.2%). Thirty-three (12.5%) had a dementia diagnosis, 74 (28.1%) a recorded carer, 19 (7.2%) lived in a care home, and 16 (6.1%) had died within 90 days of discharge. Of the 263 patients, 150 (57.0%) had medical admissions and 113 (43.0%) had surgical admissions. Emergency admissions accounted for 184 cases (70.0%), comprising 98.7% of medical and 31.8% of surgical admissions. Mean hospital stay was 9.6 days (range 1–689 days), median 3 days (IQR 1–8 days) (data not shown).
Post-discharge care utilisation
Within 90 days of discharge, 186 patients (70.7%) used healthcare services because of their index admission ‘attributable care’ (broken down by provider in Table 2). Total cost of attributable urgent and emergency care utilisation was £121 960.70; median cost of £681.00 per user (n = 52), mean £1451.91 (SD £2101.54). Total cost for attributable primary care was £13 821.66; median cost £72.79 per patient, mean £89.75 (SD £69.37) (data not shown).
Table 2. Post-discharge care utilisation (N = 263) General practice care was mostly offered on the telephone (48.6%; face-to-face, 38.0%; home visit, 9.4%; and SMS, 3.6%) by GPs (64.9%; nurse, 18.8%; pharmacist, 8.4%; allied health professionals, 4.1%; and non-clinical staff, 3.1%). Of the patients with attributable general practice care post-discharge, 120/152 (78.9%) had index emergency admissions. For 127/152 (83.6%) patients, their attributable general practice contact occurred within 14 days of discharge. No other services were used by 103/174 (59.2%) of the patients who were seen in general practice (data not shown).
Academic clinician record reviewers judged 47/263 (17.9%, 95% CI = 13.7 to 23.0) patients to have been directly attributably readmitted within 90 days of discharge, with 5/47 (10.6%) due to harms (data not shown). In univariate models (Table 3), dementia was associated with significantly increased odds of readmission (OR 2.69; 95% CI = 1.20 to 6.04; P = 0.016), as was a hospital stay of 5–14 days versus <5 days (OR 3.01; 95% CI = 1.52 to 5.99; P = 0.002) (analysis of variance: P = 0.007).
Table 3. Modelling of errors, harms, and readmissions Actions, failure to complete actions (error), and harms
Actions
Overall, 160 (60.8%; 95% CI = 54.8 to 66.6) discharge summaries requested ≥1 actions be undertaken by the general practice team, totalling 551 actionable requests. Of these discharge summaries, 128 required medication reconciliation, with 384 medication change requests (mean 1.46 requests per patient, SD 2.29), comprising: 266 requests (69.3%) to start a new repeat medication; 76 (19.8%) to discontinue; and 42 (10.9%) to change a dose. Of the 160 discharge summaries requesting actions, 45 (28.1%) requested a follow-up test or investigations, totalling 58 actionable requests (mean 1.30 per patient, SD 0.55). Referral requests appeared in 11 summaries (6.9%), and 71 (44.4%) contained other GP requests, totalling 98 actions (mean 0.37 per patient, SD 0.75) (data not shown).
Failure to complete actions (error)
In total, 50 errors (incomplete action without explanatory free text in the EHR) were identified. Most errors were medication-related (n = 30), mainly related to new repeat medication (n = 24/30) and rarely due to discontinuing (n = 2/30), changing doses (n = 2/30), or reviewing/monitoring medications (n = 2/30) (see Supplementary Table S1 for list of drugs not actioned). A total of 10 errors involved follow-up of investigations (n = 4/10 blood tests and n = 6/10 imaging/procedures). There were eight ‘other action’ errors: wound checks (n = 2/8); ReSPECT forms (n = 2/8); and physiological monitoring (n = 4/8), and two referral errors were identified. Reasons for non-completion most commonly arose from ‘clinical reasoning not documented’ actions, inappropriate or unclear requests, or missing information in discharge summaries (see Supplementary Table S2).
Of the 263 patients in the cohort, 160 had ≥1 actions. Of these patients, 33 experienced at least one error and 13 experienced multiple errors. The overall per patient error rate was 20.6% (95% CI = 15.0 to 27.6), rising to 21.5% (95% CI = 14.7 to 30.2) in patients aged ≥75 years (data not shown). Dementia (OR 4.59; 95% CI = 1.74 to 12.05; P = 0.002) and having a recorded carer (OR 2.83; 95% CI = 1.23 to 6.52; P = 0.014) were statistically significantly associated with error (Table 3). Care home residence increased odds (OR 2.60; 95% CI = 0.70 to 9.63), but not significantly (P = 0.153). Sex, ethnicity, number of GP contacts, and post-discharge follow-up were not significantly associated with error. Longer hospital stays were associated with higher odds (5–14 days, OR 2.02 and >14 days, OR 1.85), though not significantly. No errors occurred among elective admissions.
Harms
Of 50 identified errors, eight caused harm to seven patients (per patient harm rate 4.4%; 95% CI = 2.1 to 8.8). All harms occurred following emergency admissions and in patients with a recorded carer. Incidence rose to 6.5% (95% CI = 3.2 to 12.9) in patients aged ≥75 years. Of the seven harmed patients, five had directly attributable readmissions (mean stay 5.2 days; range 1–12) (data not shown). Five harms were medication related, and three test related (Table 4). Six of eight harms were graded ‘substantial’ (resulted in readmission for five patients); one ‘required intervention’ by a healthcare professional (HCP); and one was ‘potential harm’. Five harms were ‘potentially preventable’ and three ‘preventable and attributable to primary care’. Harms were significantly more likely in patients with dementia (OR 5.63; 95% CI = 1.16 to 27.19; P = 0.032) (Table 3).
Discussion
Summary
This study conducted a retrospective analysis of EHRs to examine post-discharge healthcare use, follow-up, and readmissions in patients aged ≥65 years. Post-discharge healthcare utilisation was high: nearly three-quarters of patients re-entered care within 90 days and 24% were readmitted. General practice provided follow-up for the majority, with most contacts (n = 152/174, 87.4%) attributable to the index admission. A total of 50/551 discharge actions were not completed: mostly medication reconciliation. Seven patients experienced harm directly attributable to uncompleted actions, five required hospital readmissions. Most harms were rated substantial and all were preventable or potentially preventable. Dementia and having a recorded carer increased the odds of error, harm, and readmission. Longer hospital stays were linked to higher error risk and costs. All harms followed emergency admissions.
Strengths and limitations
Including both medical and surgical admissions in the studied cohort enhanced relevance to routine general practice. Though all practices were within one region, purposive sampling captured variation in list size, setting, deprivation, and demographics. A key strength was the use of detailed EHR data, including free-text consultation notes and document attachments, enabling identification of coded and uncoded errors, linking discharge actions to care use, harms, and readmissions, and providing costing information.
A weakness common to all retrospective record analysis studies is that clinically appropriate decisions to override or defer requests may have gone unrecorded. Though harms were assessed independently using standardised tools, some subjectivity remains. The authors did not extract patient postcodes for confidentiality reasons, so individual-level deprivation could not be assessed. Due to the intensity of manual extraction, general practice contact data were limited to the first six contacts.
Comparison with existing literature
Compared with the authors’ earlier study1 in those aged ≥75 years, error rates showed a significant (53%) reduction (n = 23/107 [21.5%] versus n = 112/246 [45.5%], P<0.0001). However, secondary care action requests declined by 22% (P<0.0001) in those aged ≥75 years over this time period (64% in 2023 versus 82% in 2014), likely reflecting British Medical Association guidance to reduce workload transfer to primary care.20,21 Harm rates were not significantly different. A 2019 meta-analysis found only three of 70 harm studies were set in primary care;22 one dataset23,24 reporting a 0.5% per patient harm rate in primary care highlights the elevated risk seen in the current study’s cohort (4.4%). Failure to prescribe laxatives (traditionally considered as having low priority to HCPs) was found to be strongly associated with readmission harms in the current study, an effect previously noted in heart failure cohorts.24 With their demonstrable five-fold risk of error, people living with dementia (PLWD) are a priority for intervention after discharge.25 A meta-analysis found strong evidence for a relative risk (1.42, 95% CI = 1.21 to 1.66) for admission in PLWD,26 and cohorts in England and Wales are projected to rise to 1.2–1.7 million by 2040.27
The CQC suggests6 that the impact of premature discharge may be seen in primary care post-discharge. The current study found general practice costs to be low; secondary care costs by contrast were high with a strong right skew. Patients in the cohort did have a great deal of post-discharge general practice care; the effect of local hospitals’ discharge practices on this expenditure is yet to be studied. The current investigation of general practice costs is timely given the NHS’s ‘left shift’ agenda.14,28 Recent NHS efforts to standardise intermediate care data collection illustrate how complex and variable these secondary care pathways are, and underscore the challenges of understanding how such variation impacts the interface with general practice.29
Emerging evidence from the US and Israel suggests that when primary care providers have sufficient time to explain discharge summaries to patients, readmissions can be reduced.30,31 To replicate this in the UK, primary care must be resourced, better linked with secondary care,10 and provided with targeted tools and interventions. Given the perfect separation of error to emergency admissions in the present modelling, future focus should prioritise these discharges.
Implications for research and practice
General practices should prioritise timely review of high-risk patients, particularly emergency admissions for those with dementia or a carer. Secondary care readmission risk-prediction models should be adapted for primary care. Targeted follow-up (perhaps artifical intelligence enabled) and dedicated, funded time and space for post-discharge review would enable action, with structured tools, for example, discharge summary templates, action checklists, and coded follow-up prompts, to support consistent processing. Understanding how patient and carer communication and continuity of care influence the success of discharge follow-up in general practice is crucial. Improved linkage across primary, community, and hospital datasets may also support more detailed examination of care transitions in future research. Future interventional research should be particularly mindful of resource limitations in general practice, designed to minimise additional workload, and fit around existing literature on post-discharge multidisciplinary teams in general practice, including the NHS Additional Roles Reimbursement Scheme pharmacy roles.13
Notes
Funding
Rachel A Spencer, Annabelle Long, Zakia Shariff, and Naomi Klepacz were funded by the National Institute for Health and Care Research (NIHR) through a Career Development Advanced Fellowship awarded to Rachel A Spencer (reference: 301328). The views expressed in this publication are those of the authors and not necessarily those of the NIHR, NHS, or Department of Health and Social Care.
Ethical approval
Ethical approval was granted by East of England – Essex Research Ethics Committee (reference: 21/EE/0227). Confidential Advisory Group approval number: 23/CAG/0100. Identifiable data were accessed only within practices under this approval, and all extracted data were anonymised and stored on secure, access-restricted university servers.
Provenance
Freely submitted; externally peer reviewed.
Data
This study's bespoke dataset was created as part of the General Practice Management After Transition Events (GP-MATE) study, which explores communication between older patients/their carers and their practices after discharge. The data are not publicly available due to the sensitive nature of the pseudonymised healthcare record extraction. Please contact the corresponding author to enquire about data.
Acknowledgements
The authors would like to thank the staff at the seven general practices who shared their data and for supporting this study.
Competing interests
The authors have declared no competing interests.
