Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: A cohort study using linked primary care, secondary care, and death registration data in the OpenSAFELY platform.

BACKGROUND: There is concern about medium to long-term adverse outcomes following acute Coronavirus Disease 2019 (COVID-19), but little relevant evidence exists. We aimed to investigate whether risks of hospital admission and death, overall and by specific cause, are raised following discharge from a COVID-19 hospitalisation. METHODS AND
FINDINGS: With the approval of NHS-England, we conducted a cohort study, using linked primary care and hospital data in OpenSAFELY to compare risks of hospital admission and death, overall and by specific cause, between people discharged from COVID-19 hospitalisation (February to December 2020) and surviving at least 1 week, and (i) demographically matched controls from the 2019 general population; and (ii) people discharged from influenza hospitalisation in 2017 to 2019. We used Cox regression adjusted for age, sex, ethnicity, obesity, smoking status, deprivation, and comorbidities considered potential risk factors for severe COVID-19 outcomes. We included 24,673 postdischarge COVID-19 patients, 123,362 general population controls, and 16,058 influenza controls, followed for ≤315 days. COVID-19 patients had median age of 66 years, 13,733 (56%) were male, and 19,061 (77%) were of white ethnicity. Overall risk of hospitalisation or death (30,968 events) was higher in the COVID-19 group than general population controls (fully adjusted hazard ratio [aHR] 2.22, 2.14 to 2.30, p < 0.001) but slightly lower than the influenza group (aHR 0.95, 0.91 to 0.98, p = 0.004). All-cause mortality (7,439 events) was highest in the COVID-19 group (aHR 4.82, 4.48 to 5.19 versus general population controls [p < 0.001] and 1.74, 1.61 to 1.88 versus influenza controls [p < 0.001]). Risks for cause-specific outcomes were higher in COVID-19 survivors than in general population controls and largely similar or lower in COVID-19 compared with influenza patients. However, COVID-19 patients were more likely than influenza patients to be readmitted or die due to their initial infection or other lower respiratory tract infection (aHR 1.37, 1.22 to 1.54, p < 0.001) and to experience mental health or cognitive-related admission or death (aHR 1.37, 1.02 to 1.84, p = 0.039); in particular, COVID-19 survivors with preexisting dementia had higher risk of dementia hospitalisation or death (age- and sex-adjusted HR 2.47, 1.37 to 4.44, p = 0.002). Limitations of our study were that reasons for hospitalisation or death may have been misclassified in some cases due to inconsistent use of codes, and we did not have data to distinguish COVID-19 variants.
CONCLUSIONS: In this study, we observed that people discharged from a COVID-19 hospital admission had markedly higher risks for rehospitalisation and death than the general population, suggesting a substantial extra burden on healthcare. Most risks were similar to those observed after influenza hospitalisations, but COVID-19 patients had higher risks of all-cause mortality, readmission or death due to the initial infection, and dementia death, highlighting the importance of postdischarge monitoring.

Introduction

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) emerged in early 2020 and rapidly spread around the world, infecting >140 million people globally [1]. Acute infection can be asymptomatic or mild [2], but a substantial minority of infected people experience severe Coronavirus Disease 2019 (COVID-19) requiring hospitalisation [3], with age being a major risk factor, along with male sex, non-white ethnicity, and certain comorbidities [4-6]. Early in the pandemic, the proportion surviving hospitalisation was around 50% to 70% [7], though improved treatment guidelines and the identification of effective therapies such as dexamethasone helped to improve survival rates [8,9]. There is now a large and growing population of people who have survived a COVID-19 hospitalisation, but little is known about their longer-term health outcomes.

One recent study of United States Department of Veterans Affairs (VA) data examined a wide range of diagnoses, prescriptions, and laboratory abnormalities among 30-day survivors of COVID-19, showing excess risks of several health outcomes in the 6 months following infection, compared with the general VA population [10]. Whether these findings generalise to the entire US population or other settings remains unclear. Another US study limited to people aged <65 years also found excess risks of a range of clinical outcomes ascertained from health insurance data among people with a record of SARS-CoV-2 infection [11]. A United Kingdom study of routinely collected primary care and hospitalisation data described raised rates of all-cause hospital admission and death among patients discharged following a COVID-19 hospitalisation; the authors also noted raised risks of adverse respiratory and cardiovascular sequelae among the selected outcomes investigated [12]. Only a general population comparator was used, making it difficult to disentangle risks specific to COVID-19 from those associated with hospitalisation more generally; furthermore, a hospitalised cohort is likely to have been more prone to health problems at the outset than the general population comparator group.

Given high rates of current and past SARS-CoV-2 infection in many countries, understanding risks to health beyond acute infection is vital to support resource planning and inform measures to mitigate and reduce risks. To generate new knowledge and fill gaps in the evidence base in this important emerging area, we therefore aimed to investigate the incidence of subsequent hospital admission and death, both overall and from a wide range of specific causes, following a COVID-19 hospitalisation in England. We aimed to compare post-COVID risks with 2 separate comparison groups: (i) the general population; and (ii) people hospitalised for influenza prior to the current pandemic. The latter was included to provide a comparison with risks after hospitalisation in general, using admissions from a well-characterised infectious disease.

Methods

Study design and study population

A cohort study was carried out within the OpenSAFELY platform, which has been described previously [6]. We used routinely collected electronic data from primary care practices using TPP SystmOne software, [13] covering approximately 40% of the population in England, linked at the individual patient level to NHS Secondary Uses Service (SUS) data on hospitalisations, and Office of National Statistics (ONS) death registration data (from 2019 onwards). A brief outline study plan was created in February 2021 setting out a priori the aim and overall approach for the present study (S1 Outline Study Plan); the design was developed further in discussion with the study team prior to data analysis. We selected all individuals discharged between 1 February and 30 December 2020 from a hospitalisation that lasted >1 day and where COVID-19 was coded as the primary diagnosis (based on the International Classification of Diseases (ICD)-10 codes U07.1 “COVID-19—virus identified” and U07.2 “COVID-19—virus not identified”) and who were alive and under follow-up in a TPP practice 1 week after discharge (to avoid a focus on hospital transfers and immediate readmissions/deaths, as early descriptive data suggested a large number of outcomes in week 1 would have obscured the longer-term outcomes that were of primary interest). We excluded a small number of people with missing age, sex, or index of multiple deprivation, which are likely to indicate poor data quality. Two comparison groups were also selected. First, we identified people under follow-up in the general population in 2019, individually matched 5:1 to the COVID-19 group on age (within 3 years), sex, Sustainability and Transformation Plans (STP, a geographical area used as in NHS administration, of which there were 32 in our data), and calendar month (e.g., a patient discharged from a COVID-19 hospitalisation in April 2020 was matched to 5 individuals of the same age, sex, and STP who were under follow-up in general practice on 1 April 2019). The rationale for matching to 2019 data was to provide a comparison with routinely faced risks during prepandemic times. Second, we identified all individuals discharged from hospital in 2017 to 2019 where influenza was coded as the primary reason for hospitalisation and who were alive and under follow-up 1 week after discharge.

Outcomes and covariates

The outcomes were (i) time to first hospitalisation or death (composite outcome); (ii) all-cause mortality; and (iii) time to first cause-specific hospitalisation or death. Hospitalisations were identified from linked SUS data and included all admissions (whether planned or unplanned). All-cause mortality was identified using date of death in the primary care record so that deaths before 2019 (in the influenza group) could be included (as linked ONS data were not available prior to 2019); concordance of death dates between primary care and linked ONS data has been shown to be high [14]. Cause-specific outcomes were categorised based on ICD-10 codes into infections (ICD-10 codes beginning with “A”), cancers except nonmelanoma skin cancer (C, except C44), endocrine/nutritional/metabolic (E), mental health and cognitive (F, G30 and X60-84), nervous system (G, except G30), circulatory (I), COVID-19/influenza/pneumonia/other lower respiratory tract infections (LRTIs) (J09-22, U07.1/2), other respiratory (J23-99), digestive (K), musculoskeletal (M), genitourinary (N), and external causes (S-Y, except X60-84). For each of these, the outcome was time to the earliest of hospitalisation with the relevant outcome listed as primary diagnosis, or death with the relevant outcome listed as the underlying cause on the death certificate [15]. The influenza control group was restricted to those discharged in 2019 for analyses of these cause-specific outcomes, because we did not have linked death registration data (and thus cause of death) for earlier years.

Other covariates considered in the analysis were factors that might be associated with both risk of severe COVID-19 and subsequent outcomes, namely age, sex, ethnicity, obesity, smoking status, index of multiple deprivation quintile (derived from the patient’s postcode at lower super output area level), and comorbidities considered potential risk factors for severe COVID-19 outcomes (see Table 1 and footnotes for full specification of covariate categories and comorbidities).

Table 1

Characteristics of patients hospitalised for COVID-19 and controls.

		Hospitalised with COVID-19	Matched controls from 2019 general population	Hospitalised with influenza in 2017–2019
N (%)		24,673 (100.0)	123,362 (100.0)	16,058 (100.0)
Age (y)	18–39	2,035 (8.2)	10,175 (8.2)	2,024 (12.6)
	40–49	2,756 (11.2)	13,780 (11.2)	1,462 (9.1)
	50–59	4,679 (19.0)	23,395 (19.0)	2,126 (13.2)
	60–69	4,602 (18.7)	23,010 (18.7)	2,653 (16.5)
	70–79	5,034 (20.4)	25,170 (20.4)	3,492 (21.7)
	80+	5,567 (22.6)	27,832 (22.6)	4,301 (26.8)
	Median (IQR)	66 (53–78)	66 (53–78)	69 (52–80)
Sex	Male	13,733 (55.7)	68,662 (55.7)	7,097 (44.2)
	Female	10,940 (44.3)	54,700 (44.3)	8,961 (55.8)
BMI (kg/m 2 )	Not obese	12,710 (51.5) [54.8]	82,908 (67.2) [72.7]	10,065 (62.7) [67.4]
N (% of total)	30–34.9 (Obese class I)	5,860 (23.8) [25.3]	20,985 (17.0) [18.4]	2,853 (17.8) [19.1]
[% among nonmissing]	35–39.9 (Obese class II)	2,819 (11.4) [12.2]	7,069 (5.7) [6.2]	1,271 (7.9) [8.5]
	≥40 (Obese class III)	1,789 (7.3) [7.7]	3,015 (2.4) [2.6]	737 (4.6) [4.9]
	Missing	1,495 (6.1)	9,385 (7.6)	1,132 (7.0)
Smoking status	Never	10,350 (41.9) [42.2]	52,145 (42.3) [43.0]	5,711 (35.6) [35.8]
N (% of total)	Former	12,498 (50.7) [51.0]	52,426 (42.5) [43.2]	7,346 (45.7) [46.1]
[% among nonmissing]	Current	1,663 (6.7) [6.8]	16,699 (13.5) [13.8]	2,874 (17.9) [18.0]
	Missing	162 (0.7)	2,092 (1.7)	127 (0.8)
Ethnicity	White	19,061 (77.3) [78.3]	80,923 (65.6) [87.7]	14,035 (87.4) [88.5]
N (% of total)	Mixed	313 (1.3) [1.3]	821 (0.7) [0.9]	121 (0.8) [0.8]
[% among nonmissing]	South Asian	3,457 (14.0) [14.2]	6,727 (5.5) [7.3]	1,242 (7.7) [7.8]
	Black	920 (3.7) [3.8]	2,225 (1.8) [2.4]	251 (1.6) [1.6]
	Other	590 (2.4) [2.4]	1,572 (1.3) [1.7]	211 (1.3) [1.3]
	Missing	332 (1.3)	31,094 (25.2)	198 (1.2)
Index of Multiple Deprivation	1 (least deprived)	4,622 (18.7)	25,428 (20.6)	3,282 (20.4)
	2	4,743 (19.2)	25,259 (20.5)	3,251 (20.2)
	3	4,678 (19.0)	23,503 (19.1)	3,272 (20.4)
	4	5,183 (21.0)	24,222 (19.6)	3,133 (19.5)
	5 (most deprived)	5,447 (22.1)	24,950 (20.2)	3,120 (19.4)
Care home resident	Yes	1,197 (4.9)	1,650 (1.3)	391 (2.4)
Length of hospital stay	Median (IQR)	7 (3–13)	-	4 (2–9)
Any critical care	Yes	2,659 (10.8)	-	18 (0.1)
Comorbidities
Hypertension		12,132 (49.2)	48,565 (39.4)	7,550 (47.0)
Chronic respiratory disease		3,841 (15.6)	9,664 (7.8)	3,588 (22.3)
Asthma	With no oral steroid use	3,741 (15.2)	14,364 (11.6)	2,872 (17.9)
	With oral steroid use	1,334 (5.4)	2,375 (1.9)	1,210 (7.5)
Chronic heart disease		5,540 (22.5)	18,285 (14.8)	3,934 (24.5)
Diabetes	With HbA1c <58 mmol/mol	4,727 (19.2)	14,855 (12.0)	2,443 (15.2)
	With HbA1c > = 58 mmol/mol	3,124 (12.7)	5,567 (4.5)	1,426 (8.9)
	With no recent HbA1c measure	402 (1.6)	1,133 (0.9)	218 (1.4)
Cancer (nonhaematological)	Diagnosed <1 year ago	401 (1.6)	1,044 (0.8)	316 (2.0)
	Diagnosed 1–4.9 years ago	708 (2.9)	2,959 (2.4)	539 (3.4)
	Diagnosed ≥5 years ago	1,622 (6.6)	7,353 (6.0)	1,167 (7.3)
Haematological malignancy	Diagnosed <1 year ago	70 (0.3)	123 (0.1)	110 (0.7)
	Diagnosed 1–4.9 years ago	167 (0.7)	362 (0.3)	239 (1.5)
	Diagnosed ≥5 years ago	252 (1.0)	694 (0.6)	295 (1.8)
Reduced kidney function	Estimated GFR 30–60	4,502 (18.2)	17,986 (14.6)	3,299 (20.5)
	Estimated GFR 15-<30	481 (1.9)	1,313 (1.1)	350 (2.2)
	Estimated GFR <15 or dialysis	443 (1.8)	353 (0.3)	342 (2.1)
Chronic liver disease		414 (1.7)	901 (0.7)	222 (1.4)
Dementia		1,677 (6.8)	4,409 (3.6)	1,198 (7.5)
Stroke		1,835 (7.4)	4,275 (3.5)	1,057 (6.6)
Other neurological disease		861 (3.5)	1,817 (1.5)	574 (3.6)
Organ transplant		173 (0.7)	168 (0.1)	189 (1.2)
Asplenia		99 (0.4)	242 (0.2)	78 (0.5)
Rheum arthritis/lupus/psoriasis		2,132 (8.6)	7,717 (6.3)	1,408 (8.8)
Other immunosuppressive disease		76 (0.3)	311 (0.3)	108 (0.7)

BMI, body mass index; COVID-19, Coronavirus Disease 2019; GFR, glomerular filtration rate.

Diabetes HbA1c category was determined according to the most recent glycated haemoglobin (HbA1c) recorded in the 15 months prior to the index date; other neurological disease was defined as motor neurone disease, myasthenia gravis, multiple sclerosis, Parkinson disease, cerebral palsy, quadriplegia or hemiplegia, and progressive cerebellar disease; asplenia included splenectomy or a spleen dysfunction, including sickle cell disease; other immunosuppressive conditions was defined as permanent immunodeficiency ever diagnosed, or aplastic anaemia or temporary immunodeficiency recorded within the last year.

Information on all covariates was obtained by searching TPP SystmOne records for specific coded data, based on a subset of SNOMED-CT mapped to Read version 3 codes. Covariates were identified using data prior to the patient’s hospital admission date (for the COVID-19 and influenza groups) or the index date (for the matched control group, i.e., first day of the matched calendar month in 2019). For the COVID-19 and influenza hospitalised groups, primary care data on ethnicity were supplemented with information from the hospitalisation record, to improve completeness. We also classified individuals in residence in a care home based on address linkage; this was used in descriptive and sensitivity analyses only due to limited sensitivity [16]. All codelists, along with detailed information on their compilation are available at https://codelists.opensafely.org for inspection and reuse by the wider research community.

Statistical analysis

Follow-up began on the eighth day after hospital discharge for the COVID-19 and influenza groups, and on the first of the same calendar month in 2019 for the general population control group. Follow-up ended at the first occurrence of the analysis-specific outcome, or the earliest relevant censoring date for data availability/coverage for the outcome being analysed; the control groups were additionally censored after the maximum follow-up time of the COVID-19 group (315 days). For outcomes involving hospital admissions, the administrative censoring date (for SUS data) was 30 December 2020; for outcomes involving cause of death, the administrative censoring date (for ONS mortality data) was 11 March 2021; for the all-cause death outcome, which was ascertained in primary care data, patients were censored at date of deregistration if they had left the TPP general practice network. Cumulative incidence of the composite hospitalisation/death outcome and all-cause mortality were calculated using Kaplan–Meier methods. Hazard ratios (HRs) comparing COVID-19 and controls were estimated using Cox regression models. Separate models were fitted for the comparisons with matched 2019 general population controls (models stratified by matched set) and with influenza controls (models adjusted for age [continuous, as a 4-knot restricted cubic spline except in cause-specific outcome models where a simpler linear term was used due to lower power], sex, STP, and calendar month). The additional covariates noted above were then added to the models. Missing ethnicity was handled using multiple imputation (10 imputations) based on a multinomial logistic model including all covariates from the substantive models and an indicator for the outcome of interest; a population-calibrated multiple imputation carried out in a previous analysis in this data sources showed minimal nonrandom missingness in ethnicity data (calibration parameters were close to 0), suggesting missing at random to be a reasonable assumption [6]. People with missing data on body mass index (BMI) or smoking were excluded from regression models. Such a “complete case analysis” is valid under the assumption that missingness is conditionally independent of the outcome [17]; while this assumption cannot be verified in the data (because one cannot condition on the missing values themselves), we had no reason to doubt that recording of BMI/smoking in primary care would have been independently associated with the study outcomes; on the other hand, we deemed the missing at random assumption required for multiple imputation to be unlikely to hold for these variables in primary care (e.g., because smokers or those at the extremes of the weight distribution are more likely to have these data recorded). Cumulative incidence of cause-specific hospitalisation/death outcomes were calculated with deaths from other causes treated as a competing risk [18]. HRs for these outcomes were then estimated from a Cox model targeting the cause-specific hazard, with deaths from competing risks censored. Interactions with follow-up time (classified as <30 days, 30 to <90 days, and ≥90 days from hospitalisation [COVID-19/influenza groups] or entry [general population controls]) were examined to investigate whether any increased risk was concentrated in early follow-up and as an implicit check of proportional hazards. We also checked for proportional hazards in adjustment covariates by testing for a 0 slope in the Schoenfeld residuals for each adjusted model; where there was evidence of nonproportionality, an interaction between follow-up time and any variables with evidence of nonproportional hazards was added to the model as a sensitivity analysis. In a secondary analysis, we fitted Fine and Gray regression models to characterise overall differences in the cumulative incidence of cause-specific outcomes in the presence of competing risks. Further sensitivity analyses included restricting the COVID-19 group to those with a confirmed infection ICD-10 code (U07.1), adjusting for receipt of critical care in hospital (COVID-19 versus influenza comparison only) and adjusting for care home residence.

The study was approved by the Health Research Authority (REC reference 20/LO/0651) and by the LSHTM Ethics Board (ref 21863). An information governance statement is provided in S1 IG Statement. Data management and analysis were carried out in Python version 3.8 and Stata version 16. This study is reported according to the Reporting of Studies Conducted using Observational Routinely-Collected Data (RECORD) guideline (S1 RECORD Checklist).

Results

We included 24,673 individuals discharged after a COVID-19 hospitalisation, alongside 123,362 matched controls from the 2019 general population, and 16,058 individuals discharged after influenza hospitalisation in 2017 to 2019 (Figs 1 and S1).

Fig 1

Study flow chart.

COVID-19, Coronavirus Disease 2019; STP, Sustainability and Transformation Plans; SUS, Secondary Uses Service.

At entry, the COVID-19 group had similar age and sex distribution to the general population groups due to matching but had younger median age and were more likely to be male than the influenza group (Table 1). BMI and smoking were 93% to 99% complete in all groups; those with missing data on these variables (who were excluded from later regression modelling) were more likely to be younger, male, and from more deprived areas (S1 Table). Missing ethnicity (which was handled by multiple imputation) was <2% in the COVID-19 and influenza groups but 25% in the matched control group (no hospital-based ethnicity records were available for this group). The COVID-19 group were more likely to be obese, non-white, and less likely to be current smokers than both comparison groups. Preexisting comorbidities were more common in both COVID-19 and influenza-discharged patients than in general population controls. COVID-19 patients had longer median duration of hospital stay and were more likely to have received critical care during their admission than influenza patients.

Numbers of outcome events are shown in S2 Table. Cumulative incidence of subsequent hospital admission or death after study entry in the COVID-19 group was higher than in general population controls but slightly lower than in the influenza group (cumulative incidence at 6 months [for illustration] = 34.8%, 15.2%, and 37.8% in the 3 groups, respectively; fully adjusted hazard ratio (aHR) across all follow-up = 2.22, 2.14 to 2.30 for COVID-19 versus general population [p < 0.001]; 0.95, 0.91 to 0.98 for COVID-19 versus influenza [p-0.004], cumulative incidence curves over all follow-up shown in Fig 2A, model-specific HRs shown in Fig 3). Cumulative all-cause mortality was higher in the COVID-19 group than in both the general population and influenza groups (7.5%, 1.4%, and 4.9% at 6 months in the 3 groups, respectively; fully aHR = 4.82, 4.48 to 5.19 for COVID-19 versus general population [p < 0.001]; 1.74, 1.61 to 1.88 for COVID versus influenza [p < 0.001], Figs 2B and 3). To further explore this, causes of death were examined (S3 Table). A substantial proportion of deaths in the COVID-19 group had COVID-19 listed as the underlying cause (500/2,022, 24.7%), while in the influenza group, ≤5 deaths were coded with influenza as the underlying cause.

Fig 2

Cumulative incidence of (A) admission or death (composite outcome), and (B) all-cause mortality, in patients discharged from COVID-19 hospital admissions, influenza hospital admissions, and in matched general population controls. COVID-19, Coronavirus Disease 2019.

Fig 3

HRs comparing exposed (prior COVID-19 hospitalisation) and controls for risk of subsequent hospital admission or death (composite outcome) and all-cause mortality.

Footnotes: *All models restricted to individuals with complete data on BMI and smoking (n = 23,153/24,673 (94%) in the COVID-19 group, 113,757/123,362 (92%) in general population controls and 14,904/16,058 (93%) in influenza controls (see S1 Table). Median time at risk in the COVID-19 group was 61 days for the composite outcome and 167 days for death; total time at risk followed a bimodal distribution corresponding to the 2 main pandemic waves in England. BMI, body mass index; COVID-19, Coronavirus Disease 2019; HR, hazard ratio; IMD, index of multiple deprivation.

Cumulative incidences of cause-specific hospital admissions or deaths are shown in Fig 4. After adjustment for matching factors and other covariates, risks of all cause-specific outcomes were substantially higher in COVID-19 groups than in general population controls (Fig 5). Compared with influenza patients, people in the COVID-19 group had similar or lower risk of admission or death from most causes but higher risks of admission/death from COVID-19/influenza/LRTI (aHR 1.37, 1.22 to 1.54, p < 0.001); in the post-COVID-19 group, these outcomes were dominated by codes for COVID-19 itself (515/1,122 [46%] of hospitalisations and 342/368 [93%] of deaths) and pneumonia (461/1,122 [41%] of hospitalisations). The COVID-19 group also had higher risks than the influenza group for mental health or cognitive outcomes (aHR 1.37, 1.02 to 1.84, p = 0.039). This was further explored in a post hoc analysis of specific outcomes within the mental health and cognitive category (Table 2). Raised risks in the COVID-19 group appeared to be driven by dementia hospitalisations/deaths (age/sex-adjusted HR 2.32, 1.48 to 3.64, p < 0.001), particularly among those with preexisting dementia at baseline (HR 2.47, 1.37 to 4.44, p = 0.002) and/or resident in care homes (HR 2.53, 0.99 to 6.41, p = 0.051). Of note, 129/161 dementia outcome events (80.1%) were deaths (rather than hospitalisations). Higher rates of hospitalisations/deaths due to mood disorders and neurotic/stress-related/somatoform disorders were also observed in COVID-19 patients, but confidence intervals were too wide to be conclusive.

Fig 4

Cumulative incidence of cause-specific admission/death in patients discharged from COVID-19 hospital admissions, influenza hospital admissions, and in matched general population controls.

Footnotes: For each subpanel, the outcome was defined as the first hospitalisation or death record with an ICD-10 code in the given category listed as the primary reason for hospitalisation/underlying cause of death. Deaths from other causes were treated as competing risks. In the influenza group, only patients entering the study in 2019 were included in analysis of cause-specific outcomes, as linked cause of death data were only available from 2019 onwards. COVID-19, Coronavirus Disease 2019; ICD, International Classification of Diseases; LRTI, lower respiratory tract infection.

Fig 5

HRs comparing exposed (prior COVID-19 hospitalisation) and controls for cause-specific hospital admission/deaths.

Footnotes: In the influenza group, only patients entering the study in 2019 were included in analysis of cause-specific outcomes, as linked cause of death data were only available from 2019 onwards. All models restricted to individuals with complete data on BMI and smoking (n = 23,153/24,673 (94%) in the COVID-19 group, 113,757/123,362 (92%) in general population controls and 6,161/6,689 (92%) in influenza (2019 only) controls (see S1 Table). Median time at risk in the COVID-19 group ranged from 91 to 108 days across outcomes; total time at risk followed a bimodal distribution corresponding to the 2 main pandemic waves in England. BMI, body mass index; COVID-19, Coronavirus Disease 2019; HR, hazard ratio; LRTI, lower respiratory tract infection.

Table 2

Post hoc analysis of specific hospitalisation/mortality outcomes within the mental health and cognitive category.

	[events] rate per 1,000 person-years (95% CI)		Age- and sex-adjusted HR for COVID-19 vs influenza groups (95% CI)
	COVID-19 group	Influenza group
Dementia (F00–F03, G30)	[134] 14.74 (12.45–17.46)	[27] 5.20 (3.57–7.58)	2.32 (1.48–3.64)
(among those with baseline dementia)	[102] 168.07 (138.43–204.07)	[14] 59.72 (35.37–100.84)	2.47 (1.37–4.44)
(among those with no baseline dementia)	[32] 3.77 (2.67–5.33)	[13] 2.62 (1.52–4.52)	1.23 (0.63–2.42)
(among those resident in a care home)	[56] 127.78 (98.34–166.04)	[< = 5] 55.14 (22.95–132.47)	2.53 (0.99–6.41)
(among those not resident in a care home)	[78] 9.02 (7.22–11.26)	[22] 4.31 (2.84–6.55)	1.80 (1.08–2.98)
Delirium (F05)	[77] 8.47 (6.78–10.59)	[28] 5.39 (3.72–7.81)	1.10 (0.70–1.72)
Schizophrenia/schizotypal/delusional disorders (F20–29)	[< = 5] 0.22 (0.06–0.88)	[< = 5] 0.77 (0.29–2.05)	0.20 (0.03–1.12)
Mood disorders (F30–39)	[19] 2.09 (1.33–3.28)	[< = 5] 0.77 (0.29–2.05)	1.61 (0.54–4.79)
Neurotic/stress-related/somatoform disorders (F40–48)	[13] 1.43 (0.83–2.46)	[< = 5] 0.58 (0.19–1.79)	2.59 (0.71–9.52)
All except dementia (F05–F99)	[114] 12.54 (10.44–15.07)	[41] 7.90 (5.82–10.73)	1.12 (0.77–1.62)

COVID-19, Coronavirus Disease 2019; HR, hazard ratio.

We found evidence of changes over time in the HRs of several outcomes, with more pronounced raised risks earlier following COVID-19 hospitalisation (S2 Fig). Our results changed little in a range of sensitivity analyses, including restricting the COVID-19 group to the 21,770/24,673 (88%) with confirmed infection, adjusting for nonproportional hazards in adjustment variables, for receipt of critical care in hospital, and for care home residence (S3 Fig). In secondary analyses using Fine and Gray models, subdistribution HRs were similar to cause-specific HRs from the primary Cox models (S3 Fig).

Discussion

Patients discharged from a COVID-19 hospitalisation and surviving at least a week had more than double the risk of subsequent hospitalisation or death and a 4.8-fold higher risk of all-cause mortality than controls from the general population, after adjusting for baseline personal and clinical characteristics. Risks were higher for all categories of disease-specific hospital admissions/deaths after a COVID-19 hospitalisation than in general population controls, with excess risks more pronounced earlier in follow-up for several outcomes. Risks for most outcomes were similar or lower for people discharged from a COVID-19 hospitalisation, compared with people discharged from an influenza hospitalisation in 2017 to 2019, but the COVID-19 group had higher subsequent all-cause mortality, higher rates of respiratory infection admissions and deaths (predominantly COVID-19), and more adverse mental health and cognitive outcomes (particularly deaths attributed to dementia among people with preexisting dementia) compared with the influenza group.

Our findings are consistent with emerging evidence from early studies suggesting that a subset of people infected with SARS-CoV-2 can experience health problems for at least several months after the acute phase of their infection, with fatigue, pain, respiratory and cardiovascular symptoms, and mental health and cognitive disturbances being among the problems that have been frequently described under the term “post-acute COVID-19 syndrome” [19]; however, epidemiological characterisation of such sequelae has been limited. Small descriptive studies of COVID-19 survivors have been suggestive of high incidence of a range of outcomes including respiratory, cardiovascular, and mental health related [20,21]; the present study helps to contextualise these observations by adding explicit comparison with risks experienced by the general population and by people with a recent influenza hospitalisation.

Only a few other studies to date have compared post-COVID risks with a control group in this way. A recent study of VA data on US veterans examined a wide range of diagnostic and other outcomes in 30-day COVID-19 survivors, compared with the general VA population [10]. Among veterans where COVID-19 had led to a hospitalisation, HRs of every category of outcome were raised. This concurs with findings from our study, despite different characteristics of the VA population. In the UK, an earlier study found an 8-fold higher risk of death in post-acute COVID-19 patients compared with general population controls, and raised risks of respiratory disease, diabetes, and cardiovascular disease [12]. Interestingly, recent data from Denmark suggest limited postacute complications following nonhospitalised COVID-19 [22]; this is in contrast to a recent study using US health insurance data, which found raised risks of a range of outcomes among a relatively young cohort with mostly (92%) nonhospitalised COVID-19 disease, compared with both the general population and people with a record of other viral LRTIs [11].

Our data showed that COVID-19 hospitalised patients were more likely to have baseline comorbidities than general population controls, reflecting known associations between comorbidities and risks of severe COVID-19 outcomes [6]. Differences in outcomes between hospitalised patients and general population controls might therefore reflect baseline differences not fully captured in our adjustment models and might also reflect a generic adverse effect of hospitalisation [23]. This is supported by the more similar risks we observed when COVID-19 survivors were compared with people who had experienced influenza hospitalisation, with risks for some outcomes actually lower in the COVID-19 group, possibly linked to a general reduction in health seeking for non-COVID conditions in the early months of the pandemic [24]. However, all-cause mortality was substantially higher after COVID-19 compared with influenza. A quarter of deaths after a COVID-19 hospitalisation had COVID-19 listed as the underlying cause, but it is not clear from our data whether patients experienced specific complications after hospital discharge that were then attributed to COVID-19, and the possibility of persistent viraemia in these patients cannot be excluded from our data. It is possible that high levels of awareness of COVID-19 during the pandemic may have encouraged coding of subsequent deaths as COVID-19-related, leading to overestimation in the comparison with historical influenza hospitalisations.

Our analysis of cause-specific outcomes also suggested a disproportionate rate of dementia deaths post-COVID-19, particularly among those with preexisting dementia. Cognitive decline after hospitalisation and critical illness have been previously described [25,26]; acute COVID-19 and associated hospital admission, social isolation, and medications may have accelerated progression of patients’ dementia; it is unclear whether postdischarge care was adequate for this vulnerable group. However, it is possible that deaths where the underlying cause was recorded as dementia may have been due to progression of underlying health problems following an acute illness as well as difficulty in managing these due to dementia. COVID-19–related delirium may have also triggered or worsened emerging dementia in some patients, or even driven a degree of misclassification given the potential clinical challenge in distinguishing between subacute or chronic delirium and progressive dementia. Due to small numbers, we could not confirm whether higher rates of mood disorders and neurotic/stress-related/somatoform disorders after COVID-19 compared with influenza were due to chance, but a number of previous studies outside the pandemic context have found that critical illness is associated with raised risks of depression, anxiety, and posttraumatic stress [27-29]. It will be important to continue to monitor these outcomes as more follow-up accumulates.

We identified COVID-19 hospitalisations and controls from a base population based on English primary care records. Around 98% of the population are registered with a general practice [30], minimising selection biases due to health-seeking behaviours, and our data source covered around 40% of the population of England, giving us high statistical power, though it should be noted that our study population would not have been geographically representative of England, since TPP SystmOne software is more widely used than other systems in parts of Eastern and Southern England and used less than other software in London [13]. We examined a broad range of hospitalisation and mortality outcomes and were able to describe and adjust for a wide range of personal and clinical characteristics using rich primary care data. Our findings were robust in a range of sensitivity analyses.

However, our study has some limitations. We relied on ICD-10 codes entered as the primary reason for hospitalisation or underlying cause of death to define our cause-specific outcomes, but these fields may not have been used consistently [31]. In particular, there might have been a tendency for clinicians aware of a recent COVID-19 hospitalisation to code COVID-19 for a range of clinical complications, masking more specific sequelae. Outcomes were classified in broad categories to obtain an overview of post-COVID-19 disease patterns; more granular disease categories would be of future interest but will require more follow-up to maintain statistical power. Our main comparisons may have been affected by time-related factors. We compared post-COVID patients in 2020 with controls from 2019 and earlier; consultations for non-COVID-19 conditions in 2020 are known to have been subdued in the general population [24], perhaps due to lockdown or public reluctance to seek care, potentially affecting comparison with earlier years. On the other hand, patients with a recent COVID-19 hospitalisation may assume immunity from reinfection and be less reticent in seeking care than the general population. The comparison with influenza may also have been affected by seasonality, since the first wave of COVID-19 in England happened outside the typical influenza season. Lack of overlap in the data meant that we could not incorporate seasonal adjustment into our statistical models for this comparison; any confounding by seasonality is likely to have led to underestimation of HRs comparing COVID-19 and influenza patients, since cases of the former were underrepresented in the winter months (which typically confer higher health risks). We had no data on whether influenza hospitalisations were confirmed by PCR testing, raising the possibility of misclassification in this comparator, though we only included cases where influenza was coded as the primary reason for hospitalisation. We did not have detailed data on disease severity, though descriptive data showed that COVID-19 patients tended to have longer hospital stays and more critical care than those hospitalised for influenza. Data were also unavailable on new/emerging COVID-19 variants during the study period. COVID-19 patients in our study had to survive at least a week to enter the study, so our results will not capture the total public health burden from point of discharge given a substantial number of deaths and readmissions in the first week following discharge; however, we felt that excluding this first week enabled a focus on medium and longer-term postacute outcomes and avoided our results being dominated by deaths and readmissions driven by premature discharge and transfers to other hospitals. Our analysis of cause-specific outcomes made an assumption of independent censoring, but deaths from competing outcomes were censored and may have been related to risk of the outcomes under study; our results are likely to have been robust to some violation of independence because the proportion of patients censored due to death from other causes was low (ranging from 1.4% to 2.6% of the study population for specific analyses). Fine and Gray modelling (which does not censor competing events) showed a similar pattern of results to the primary analysis.

Patients surviving a COVID-19 hospitalisation for at least a week after discharge were at substantially higher risk than the general population for a range of subsequent adverse outcomes over a period of up to 10 months’ follow-up included in this study. Risks for most outcomes were broadly comparable to those experienced by influenza hospitalisation survivors prior to the pandemic, but in the period following hospital discharge, COVID-19 patients had higher risks of all-cause mortality, readmission or death attributed to their initial infection, and adverse mental health and cognitive outcomes; in particular, among people with preexisting dementia, we observed an excess of deaths where dementia was recorded as the underlying cause. These findings suggest a need for services to support and closely monitor people following discharge from hospital with COVID-19, for example, through more frequent/active follow-up in primary care in the weeks and months following a hospitalisation. Our results can be used to help inform healthcare providers and raise awareness of potential complications during this period. Our findings will also help with public health resource planning in the context of high rates of SARS-CoV-2 infection in many countries. Ongoing monitoring will be important to investigate whether these patterns persist in the light of new variants and increasing levels of vaccination.

Patient and public involvement

Patients were not formally involved in developing this specific study design that was developed rapidly in the context of a global health emergency. We have developed a publicly available website (https://opensafely.org/) through which we invite any patient or member of the public to contact us regarding this study or the broader OpenSAFELY project.

The views expressed are those of the authors and not necessarily those of the NIHR, NHS England, Public Health England, or the Department of Health and Social Care.

Original priori study plan created in February 2021, and list of changes with justification.

(PDF)

Click here for additional data file.

Information Governance Statement.

(PDF)

Click here for additional data file.

Completed REporting of studies Conducted using ObseRvational Data (RECORD) checklist.

(PDF)

Click here for additional data file.

Distribution of entry dates for those in the COVID-19 hospitalised group and the influenza-hospitalised and matched general population comparison groups.

COVID-19, Coronavirus Disease 2019.

(PDF)

Click here for additional data file.

Changes over time in the HRs comparing outcomes in the COVID-19 and control groups.

COVID-19, Coronavirus Disease 2019; HR, hazard ratio; LRTI, lower respiratory tract infection.

(PDF)

Click here for additional data file.

aHRs/sHRs in sensitivity analyses.

aHR, adjusted hazard ratio; COVID-19, Coronavirus Disease 2019; HR, hazard ratio; sHR, subdistribution hazard ratio.

(PDF)

Click here for additional data file.

Demographic characteristics of people excluded from complete case analyses due to missing obesity or smoking data.

(PDF)

Click here for additional data file.

Distribution of first outcomes (hospital admission or death) among included individuals.

(PDF)

Click here for additional data file.

Leading causes of death in COVID-19 and influenza groups.

(PDF)

Click here for additional data file.

15 Jun 2021

Dear Dr Bhaskaran,

Thank you for submitting your manuscript entitled "Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: cohort study in OpenSAFELY using linked primary care, secondary care and death registration data" for consideration by PLOS Medicine.

Your manuscript has now been evaluated by the PLOS Medicine editorial staff and I am writing to let you know that we would like to send your submission out for external peer review.

However, before we can send your manuscript to reviewers, we need you to complete your submission by providing the metadata that is required for full assessment. To this end, please login to Editorial Manager where you will find the paper in the 'Submissions Needing Revisions' folder on your homepage. Please click 'Revise Submission' from the Action Links and complete all additional questions in the submission questionnaire. Please also include line numbers in your re-submitted manuscript.

Please re-submit your manuscript within two working days, i.e. by Jun 17 2021 11:59PM.

Login to Editorial Manager here: https://www.editorialmanager.com/pmedicine

Once your full submission is complete, your paper will undergo a series of checks in preparation for peer review. Once your manuscript has passed all checks it will be sent out for review.

Feel free to email us at plosmedicine@plos.org if you have any queries relating to your submission.

Kind regards,

Louise Gaynor-Brook, MBBS PhD

Associate Editor

PLOS Medicine

8 Sep 2021

Dear Prof. Bhaskaran,

Thank you very much for submitting your manuscript "Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: cohort study in OpenSAFELY using linked primary care, secondary care and death registration data" (PMEDICINE-D-21-02562R1) for consideration at PLOS Medicine.

Your paper was evaluated by four independent reviewers, including a statistical reviewer, and was discussed among all the editors here and with an academic editor with relevant expertise. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a revised version that addresses the reviewers' and editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response, and we plan to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by Sep 29 2021 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Louise Gaynor-Brook, MBBS PhD

Associate Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

General comments:

Generally, the style should be "We included 24,673 post-discharge COVID-19 patients ...", for example; numbers should be written out at the start of sentences

Throughout the manuscript, please avoid inadvertent causal language e.g. “associated with higher risk” rather than “COVID-19 patients had higher risk...”

Throughout the paper, please adapt reference call-outs to the following style: "... every year [1,2]." (noting the absence of spaces within the square brackets).

Data availability:

PLOS Medicine requires that the de-identified data underlying the specific results in a published article be made available, without restrictions on access, in a public repository or as Supporting Information at the time of article publication, provided it is legal and ethical to do so. If the data are not freely available, please describe briefly the ethical, legal, or contractual restriction that prevents you from sharing it. Please also include contact information for data request enquiries (web or email address), noting that a study author cannot be the contact person for the data.

Title: Please revise your title according to PLOS Medicine's style. Please place the study design in the subtitle (ie, after a colon). We suggest “Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: a cohort study using linked primary care, secondary care and death registration data in the OpenSAFELY platform” or similar

Abstract Methods and Findings:

Please provide brief demographic details of the study population (e.g. sex, age, ethnicity, etc)

Please include the study design in your Abstract.

Please include the important dependent variables that are adjusted for in the analyses.

Please specify that aHRs are presented with 95% CIs

Please provide numerical results to substantiate that “COVID-19 survivors with pre-existing dementia had higher risk of dementia death”

In the last sentence of the Abstract Methods and Findings section, please describe 2-3 of the main limitation(s) of the study's methodology.

Abstract Conclusions:

Please begin your Abstract Conclusions with "In this study, we observed ..." or similar, to summarize the main findings from your study without overstating your conclusions. Please address the study implications substantiated by the results.

Author Summary:

At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary

In the final bullet point of ‘What Do These Findings Mean?’, please describe the main limitations of the study in non-technical language.

Introduction:

Please indicate whether your study is novel and how you determined that, being careful to temper assertions of primacy.

Methods:

Did your study have a prospective protocol or analysis plan? Please state this (either way) early in the Methods section. If a prospective analysis plan (from your funding proposal, IRB or other ethics committee submission, study protocol, or other planning document written before analyzing the data) was used in designing the study, please include the relevant prospectively written document with your revised manuscript as a Supporting Information file to be published alongside your study, and cite it in the Methods section. A legend for this file should be included at the end of your manuscript. If no such document exists, please make sure that the Methods section transparently describes when analyses were planned, and if/when reported analyses differed from those that were planned. Changes in the analysis-- including those made in response to peer review comments-- should be identified as such in the Methods section of the paper, with rationale. If a reported analysis was performed based on an interesting but unanticipated pattern in the data, please be clear that the analysis was data-driven.

Please ensure that the study is reported according to the RECORD guideline, and include the completed RECORD checklist as Supporting Information. Please add the following statement, or similar, to the Methods: "This study is reported as per the REporting of studies Conducted using Observational Routinely-collected Data (RECORD) guideline (S1 Checklist)." The RECORD guideline can be found here: https://www.record-statement.org/ When completing the checklist, please use section and paragraph numbers, rather than page numbers which will likely no longer correspond to the appropriate sections after copy-editing.

Results:

Please incorporate Figure A1 into the main paper.

Where adjusted analyses are presented, please also provide the unadjusted analyses, and indicate which factors are adjusted for (e.g. please list comorbidities) in the tables/figure legends.

Discussion:

Please present and organize the Discussion as follows: a short, clear summary of the article's findings; what the study adds to existing research and where and why the results may differ from previous research; strengths and limitations of the study; implications and next steps for research, clinical practice, and/or public policy; one-paragraph conclusion.

Please remove all subheadings within your Discussion e.g. Key findings

Tables:

Please define abbreviations used in the table legend of each table e.g. GFR (Table 1)

Table A1 - please specify what is represented by numbers in brackets ( )

References:

Please ensure that journal name abbreviations match those found in the National Center for Biotechnology Information (NCBI) databases, and are appropriately formatted and capitalised.

Please also see https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references for further details on reference formatting.

Supplementary files:

Please see https://journals.plos.org/plosmedicine/s/supporting-information for our supporting information guidelines.

Comments from the reviewers:

Reviewer #1:

Thanks for the opportunity to review your manuscript. My role is a statistical reviewer so my comments focus on the design, analysis, and data in this manuscript. I've put overall comments first followed by queries relevant to a specific section of the manuscript (with a line reference).

This is an interesting and important study. Routinely collected health data was used to create a cohort that enable a comparison of medium-term outcomes in patients hospitalised for COVID-19, to hospitalised influenza patients and general population (patients registered in the GP system). The COVID patients were directly matched to the comparators based on age (within 3 years), sex, area, and calendar month. Three outcomes were examined - all-cause mortality/first hospitalisations, all-cause mortality, and cause-specific hospitalisation. COVID patients had higher risk of the composite outcome than general population, but similar rates to the influenza in-patients. COVID patients had higher mortality than the other groups, and similar risk of cause-specific apart from subsequent infection or other respiratory infection and mental health/cognitive related admissions/deaths.

The manuscript is well-written and clearly described (particularly given the complexities of assembling and analysing the data). As a side-note the fact that it is possible to do this study so quickly shows the value of setting up the capacity for linked health data to be used in research and is very impressive (I am jealous you have such timely access to this data).

The GP records were used to create the 'general population group' - is it effectively universal for everyone to be on this database, is there the possibility that this could miss patients who can't/don't seek healthcare?

L106. The use of the general population comparator is clear, perhaps worth clarifying here in the aims that the influenza hosp. population addresses the issue of potential for increased risk of events after any hospital exposure.

L114. Is the 40% sample broadly representative? Do some regions contribute more than expected than from their relative population levels?

L128. What was the rationale of matching to the same calendar date a year earlier rather than the same year? Was this because of availability of linked mortality data? This does make the assumption that overall mortality and hospitalisations rates of those not affected by COVID or influenza were similar in 2019 to 2020 - is there any information you can provide that supports this?

L129. Are the ICD-10 codes used here the same as presented in Table A1?

L134. Were the hospitalisations used to derive the outcomes 'acute', e.g. would this outcome include in-patient episodes for planned procedures like cataract surgery or just unplanned admissions?

L164. Why was the 8th day after discharge used for COVID and Influenza patients? Also, do influenza patients have similar in-hospital stays to the COVID patients? E.g. time spent in ICU, time on mech ventilation, overall LOS?

L174. The K-M graphs seem to show proportional hazards for comparisons between the key groups, how was this assumption checked during the analysis for these, and also the other covariates (and the form of any continuous covariates in the analysis)?

L177. How was age included in the model, as a categorical variable or continuous?

L180. This is certainly true, but there hasn't been any information presented to demonstrate that MAR is a likely mechanism of missingness. The number excluded from each of these models should be presented, and with a comparison of characteristics by missingness. This does also make the interpretation of the difference between the adjusted and non-adjusted models more difficult as they are being done on different samples. In the previous publication using OpenSAFELY, an imputation model was used to handle the missing ethnicity - could this be applied for these analyses as well?

L185. What software was used for these analyses and data management?

L185. Censoring for death assumes that the cause-specific event would be independent of mortality, which seems unlikely. Typically for this type of analysis I would use a Fine-Grey model rather than K-M, although the interpretation of effect estimates from F-G is little more complicated than K-M. Would you consider that censoring is independent from the hospitalisations here, and did you do any sensitivity analyses to check if the K-M estimator is reasonably here?

L263. Using influenza is clever but it does assume that the hospital stay for influenza and COVID patients is similar - I think this is something that should be acknowledged or some more context provided so it's clear how comparable these patients are.

Figure A1. There is small cell suppression on the consort flow-chart (second box of exclusions) - was this a condition of the accessing the data? It's not something I'm used to, but I can still make sense of what is happening and reconcile the numbers so I think this is ok.

Figure 1. These figures are ok for review purposes but they could be smartened up a bit, e.g. directly labelling the cum incidence curves instead of the legend would make it easier to read (the legend takes up ~25% of the plot area), maybe colour in addition to line type, label on the Y axis

Figure 2. Again - perfectly readable, but maybe a few tweaks (e.g. x axis goes out to 20 when there doesn't seem to be any CIs that reach that level), also the null effect line overwrites some of the labels.

Reviewer #2: Rosie Cornish

This study provides important evidence about outcomes following hospitalisation due to COVID-19. I only have a few comments.

1. The authors have used a complete case analysis, which is justified as being appropriate provided missingness is conditionally independent of the outcome. Did the authors investigate this? It seems like a plausible assumption, but it would be helpful to state somewhere whether the data supported this.

2. Care home status is listed in Table 1 but it is not mentioned as a variable included in the fully adjusted models. Since this is a factor strongly associated with COVID-19 as well as the outcomes, I think it should be included as a covariate.

3. In the methods the authors state that COVID-19 discharges were matched on age, sex, geography and time (month). It might therefore be helpful to explain in the text why the distribution of entry dates for COVID-19 discharges does not match the distribution among the general population controls as shown in Figure A2.

4. It would be helpful to indicate on Figures 2 and 4 the numbers included in each analysis so readers can see how the sample sizes for the different models differ.

5. In the results section the authors state that people in the COVID-19 group had similar or lower risk of admission or death from most causes, whereas in the abstract they say that most risks were similar (i.e. "or lower" is omitted). I think it is important to state this in the abstract as well as in the results.

6. Figure 1 states that there were 79,502 matched controls but a figure of 123,362 is given elsewhere.

7. The final point on Figure 1 states "(6,689 were discharged in 2019 and had linked data registration available)". I think this should read "… and had linked death registration data available".

Reviewer #3: This study used routinely collected health administrative data covering 40% of England's population to examine medium to long-term hospitalization and mortality following acute COVID-19. They included 24,673 patients discharged between February and December 2020 who survived 8 days without readmission. Comparison groups were: (i) matched patients in the general population in 2019, and (ii) patients discharged from hospital with influenza (2017-2019). Although incidence of subsequent hospitalisation or death in the COVID-19 group was higher than in general population controls, incidence was comparable to the influenza group (34.8%, 15.2%, and 37.8%, respectively). Compared to the influenza group, the COVID-19 group had higher all-cause 6-month mortality (7.5% vs. 4.9%), and greater risk of death for patients with dementia (HR 2.32). This study demonstrates the importance of selecting appropriate control groups and presents a new finding on the association between COVID-19 and dementia.

This paper was well-written and quite succinct. The topic of long-term outcomes after COVID-19 is highly topical and important given how many people have had COVID-19 infection. The inclusion of a hospitalized influenza control group (from before the pandemic) was appropriate, and consistent with other research (Verma AA et al. CMAJ March 22, 2021 193 (12) E410-E418).The stratified analyses around the dementia outcome is appreciated- this is helpful to address possible confounding related to care home residence.

Major comments:

The authors should further elaborate on the relevance of the primary research question - the introduction and discussion should make a case for why this study needed to be done, beyond just "to strengthen the evidence base" (line 102). Furthermore, the discussion needs further detail on how the results might be useful, and what further research would be most valuable here.

Further discussion is also warranted about why COVID patients had more COVID readmissions in the context of published data on long COVID symptoms as well as persistence of viremia in some cases. These are possibly important differences with influenza that explain the findings.

The findings with respect to dementia are interesting however what is missing is discussion of the relationship between COVID, delirium and dementia. We know that delirium episodes often worsen dementia, and that it is also often difficult to distinguish clinically between subacute/chronic delirium and progressive dementia. Hence it is quite possible that the signal related to dementia outcomes reflects COVID-related delirium (which is well described), either directly or indirectly. The other important consideration is that those who were diagnosed with COVID were isolated (presumably to a greater extent than those with influenza), and a lack of interpersonal contact is also known to worsen dementia/delirium outcomes.

Other suggestions for improvement:

Introduction

- the middle paragraph has a lot of information about other literature which would be better placed in the discussion- I suggest reducing the amount of detail in here and instead making a stronger case for why this research needed to be done.

Methods

- The authors point to a previous paper describing the OpenSAFELY program, but I think some more details on this are needed. For example, which 40% of the population does it cover? Is this percent of the population generalizable to the other 60% of the population? I ask this because the overall sample appears low (108,308) compared to the overall number of people who have been diagnosed with COVID-19 in England

- what does "under follow-up" mean? (ln. 120)- does this just mean they had not yet experienced an outcome? Please clarify.

Results

- what was the median, and minimum observation time for outcomes across patients? I see that the maximum was 315 days. This question also relates to what "under follow-up" means- how complete was the follow-up? Since this sounds like health administrative data, my expectation was that one would only be lost to follow-up if they moved away, but this should be clarified.

- tables and figures were very helpful and nicely done

- I note that the COVID group experienced fewer cancer-related outcomes than the influenza group, however this was not mentioned in the results text. I think this finding is worthy of some discussion (for discussion section as well)- could this be because the COVID group had lower rates of cancer at baseline (possibly due to extra precautions taken by cancer patients to reduce their exposure?).

Reviewer #4: "Overall and cause-specific hospitalization and death after COVID-19 hospitalization in England: cohort study in OpenSAFELY using linked primary care, secondary care and death registration data" (manuscript ID: PMEDICINE-D-21-02562R1)

SUMMARY: This cohort study of over 164,000 hospitalized adults (n=24,673 discharged following COVID-19 between discharged between 1st February and 30th December 2020; n=123,362 matched general population controls in 2019; and n=16,058 discharged following influenza between 2017-2019) compares the medium and long-term risks of hospital admission and death, overall and by specific cause across the three study groups. It uses administrative data sources from linked primary care and hospital data in the OpenSAFELY platform, which is reported to cover approximately 40% of the population in England. The main finding identifies that people discharged following hospitalization for COVID-19 had higher associated risks for rehospitalization and death than the general population, similar risks compared to those hospitalized for influenza.

The paper is well written and addresses an evolving, poorly studied, and important area of health policy and planning as it relates to the care of patients who survive hospitalization for COVID-19. Early data indicate these individuals appear to be at high risk for ongoing health needs and high resource use. The present study appears to build on a recent study in the UK (Ayoubkhani D et al BMJ 2021) by including an "active control" population of adults hospitalized with influenza and with a longer study follow-up. The study has clear applications to healthcare resource planning and policy in the care of these individuals suggesting a substantial extra burden on healthcare systems in the future.

I have a few concerns with the study as it stands that I believe should be addressed to improve the overall quality of this paper, which I hope are viewed as helpful.

1) The authors used Cox regression models to estimate hazard ratios for the defined outcomes. However, a key assumption of this modelling approach is that these HRs are constant over time. Prior literature on risk of death or rehospitalization demonstrates higher risk in the early period following discharge from hospital, as does Figure 1a provided by the authors. It would be helpful if the authors included time as an interaction term in their modelling to help delineate how these risks change over time, including the possibility of categorizing the time variable into clinically relevant post-discharge time frames (e.g. 30-, 90-, 180-day readmission).

2) Why did the authors choose not to adjust for admission to ICU in their modelling as a surrogate for disease severity? There were nearly 11% of COVID-19 patients and only 0.1% of influenza patients admitted to ICU. This may serve to substantially overestimate the adjusted risks of mortality among patients with COVID-19. Further focus on this could be added to the limitations in the discussion section as well.

3) Related to #2 above, why did the authors not adjust for readmission risk using one of the many available readmission risk indices (see Kansagara D et al JAMA 2011 for a comprehensive list)?

4) The cause-specific outcomes among adults with COVID-19 may be artificially higher than those with Influenza due to availability bias. Put simply, patients and providers may be much more aware of COVID-19 and its complications, including those related to return to hospital than might be the case for those with pneumonia or even confirmed influenza. This appears to be supported by the data presented in this study. For example, the authors report that, "A substantial proportion of deaths in the COVID-19 group had COVID-19 listed as the underlying cause (500/2022, 24.7%), while in the influenza group ≤5 deaths were coded with influenza as the underlying cause"; and "Compared with influenza patients, people in the COVID-19 group had similar or lower risk of admission or death from most causes, but higher risks of admission/death from COVID-19/influenza/lower respiratory tract infection (LRTI, adjusted HR 1.37, 1.22-1.54); in the post-COVID-19 group these outcomes were dominated by codes for COVID-19 itself (515/1122 [46%] of hospitalizations and 342/368 [93%] of deaths)". I worry that the conclusions drawn from this study in relation to the influenza comparator group may be different when considering the lack of adjustment for disease severity and this potential source of bias.

5) It is unclear as to why the authors chose to report cumulative incidence of the outcomes at 6 months when the maximum duration of follow-up for COVID-19 patients is 315 days (and the general population group was censored using this time frame). Is this the median follow-up time for the cohort? Further clarification as to why this specific time point was chosen would be informative.

6) The authors compare baseline characteristics of the three study groups and report for example that, "The COVID-19 group were more likely to be obese, non-White and less likely to be current smokers than both comparison groups". It would be helpful to report standardized differences in Table 1 to compare the magnitude of differences between study groups, which may inform some of the residual confounding that may be unaccounted for.

7) There is a risk of misclassification of hospitalization for COVID using ICD-10 code U07.2 "COVID-19 - virus not identified". To test the robustness of their findings, could the authors repeat a sensitivity analysis among those with only the ICD-10 code "U07.1 COVID-19 - virus identified"?

8) I am concerned about the presence of immortal time bias in requiring that adults survive 7 days post discharge, which also serves to underestimate the magnitude of healthcare resource needs among adults who survive hospitalization for COVID-19. There were ~3,700 patients who died within 7 days of discharge or were transferred to hospice (where they presumably died), which is ~15% of the final cohort of adults hospitalized with COVID-19 (n=24,673). The authors should mention this in their limitations and modify the interpretation of the study to include that this applies to adults who survive at least 1 week following hospitalization for COVID-19.

8) How confident are the authors in the accuracy of hospital coding for influenza admissions? Are these hospital coding practices based on PCR confirmed influenza?

9) Can the authors comment on the generalizability of the study in using the OpenSAFELY data? Specifically, there are considerations about the generalizability to the overall study cohort, and to those that acquired COVID. The low proportion of patients residing in care homes (4.9% in the COVID population and 1.3% in the matched general population) may suggest this is a healthier, less frail population overall.

10) I believe the paper could be improved with a more substantive discussion of the policy applications for health and human resource planning in the care of patients who survive hospitalization for COVID-19 beyond, "These findings suggest a need for services to support and closely monitor people following discharge from hospital with COVID-19." For example, how might these patients require more support? Will it require more frequent follow-up from their General Practitioners? Does the healthcare system need to provide education to healthcare providers to improve their comfort and competency to deal with the potentially unique needs of these patients? Should governments consider financial motivations to incentives the care of this vulnerable population?

Any attachments provided with reviews can be seen via the following link:

[LINK]

11 Oct 2021

Submitted filename: plos med responses v1.1.docx

Click here for additional data file.

8 Nov 2021

Dear Dr. Bhaskaran,

Thank you very much for re-submitting your manuscript "Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: a cohort study using linked primary care, secondary care and death registration data in the OpenSAFELY platform" (PMEDICINE-D-21-02562R2) for consideration at PLOS Medicine.

I have discussed the paper with our academic editor and it was also seen again by three reviewers. I am pleased to tell you that, provided the remaining editorial and production issues are fully dealt with, we expect to be able to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We hope to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript.

Please note, when your manuscript is accepted, an uncorrected proof of your manuscript will be published online ahead of the final version, unless you've already opted out via the online submission form. If, for any reason, you do not want an earlier version of your manuscript published online or are unsure if you have already indicated as such, please let the journal staff know immediately at plosmedicine@plos.org.

Please let me know if you have any questions, and we look forward to receiving the revised manuscript.

Sincerely,

Richard Turner PhD, for Louise Gaynor-Brook, MBBS PhD

Senior Editor, PLOS Medicine

rturner@plos.org

------------------------------------------------------------

Requests from Editors:

Please trim the data access statement. If OpenSAFELY does not currently permit access consistent with PLOS' data policy (https://journals.plos.org/plosmedicine/s/data-availability) please state that and briefly explain the reason(s).

At line 47, please make that "February to December".

At line 53, for example, please use the general style "13,733").

At line 54, please make that "adjusted Hazard Ratio [aHR] 2.22 ..." and "aHR" can be used thereafter in the abstract.

At line 55, should that be "aHR 0.95"?

At line 70, please make that "readmission or death" or similar.

In the abstract and throughout the text, please quote p values alongside 95% CI, where available.

At line 307 and any other instances, please avoid "nearly 5-fold" in favour of quoting the actual value.

Please remove the information on data availability, funding and competing interests from the end of the main text. In the event of publication, this information will appear in the article metadata, via entries in the submission form.

In the reference list, please correct the citations for references 10 and 11.

Please ensure that journal names are abbreviated consistently (e.g., "Lancet" and derivatives rather than "The Lancet" etc).

We suggest breaking the analysis plan out into a separate attached file labelled "S1_Analysis_Plan" or similar, referred to in the Methods section (main text).

Please adapt the label for the RECORD checklist to "S2_RECORD_Checklist" or similar, and refer to it by this name in the Methods section.

Comments from Reviewers:

*** Reviewer #1:

Thanks for the revised manuscript and replies to my original queries. The responses are comprehensive and apart from one small detail (in the flow diagram) I think this manuscript looks excellent and I recommend that it is accepted. I appreciate the work you have done in this revision - it makes the entire body of work robust.

The GP coverage information is helpful - I think acknowledging this as a limitation is all that's needed here. The additional details and tests about the proportional hazards are appropriate. The additional details and modified analysis (i.e. the ethnicity adjusted models) are appropriate and meet my initial query. The sensitivity analyses using Fine and Grey are a good addition and it's reassuring to see similar results to the Cox models.

Only one small check for Figure 1 - Is the number in the flowchart that died during hospitalisation or transfer (n=30,070) correct? This doesn't match the number before and after this exclusion was applied (108,308/75840).

*** Reviewer #3:

Thank you for addressing my comments.

My only remaining concern relates to the dataset coverage of "OpenSafely": in addition to mentioning data coverage as a limitation in the discussion section, the details of which 40% of the population are covered should be included in the Methods section as well.

We identified a few minor typographical errors in the "Why was this study done" section- line 86 "to compared", line 100 "by increased".

*** Reviewer #4:

The authors should be commended for addressing all of our concerns raised during review, including completing several additional analyses to test the robustness of their findings and adding an expanded discussion of the study limitations and policy applications. As a result, we believe that these steps served to substantially strengthen their conclusions which greatly improved the overall quality of the manuscript. Congratulations on completing this important work!

***

Any attachments provided with reviews can be seen via the following link:

[LINK]

11 Nov 2021

Submitted filename: editor and reviewer responses at R3.docx

Click here for additional data file.

17 Nov 2021

Dear Dr Bhaskaran,

On behalf of my colleagues and our Academic Editor, Dr Basu, I am pleased to inform you that we have agreed to publish your manuscript "Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: a cohort study using linked primary care, secondary care and death registration data in the OpenSAFELY platform" (PMEDICINE-D-21-02562R3) in PLOS Medicine.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. Please be aware that it may take several days for you to receive this email; during this time no action is required by you. Once you have received these formatting requests, please note that your manuscript will not be scheduled for publication until you have made the required changes.

Prior to final acceptance, please:

Adapt "hospitalisation/death", and similar forms of words in the abstract, to "hospitalisation or death" as appropriate;

Remove "nearly" at line 308; and

Trim the section on "Information and governance" to no more than 5 lines, and move it to the Methods section (some of the information is already present in that section and does not need to be duplicated).

In the meantime, please log into Editorial Manager at http://www.editorialmanager.com/pmedicine/, click the "Update My Information" link at the top of the page, and update your user information to ensure an efficient production process.

PRESS

We frequently collaborate with press offices. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximise its impact. If the press office is planning to promote your findings, we would be grateful if they could coordinate with medicinepress@plos.org. If you have not yet opted out of the early version process, we ask that you notify us immediately of any press plans so that we may do so on your behalf.

We also ask that you take this opportunity to read our Embargo Policy regarding the discussion, promotion and media coverage of work that is yet to be published by PLOS. As your manuscript is not yet published, it is bound by the conditions of our Embargo Policy. Please be aware that this policy is in place both to ensure that any press coverage of your article is fully substantiated and to provide a direct link between such coverage and the published work. For full details of our Embargo Policy, please visit http://www.plos.org/about/media-inquiries/embargo-policy/.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Thank you again for submitting to PLOS Medicine. We look forward to publishing your paper.

Sincerely,

Richard Turner PhD, for Louise Gaynor-Brook, MBBS PhD

Senior Editor, PLOS Medicine

rturner@plos.org