Open access Original research

Evaluating access to health and care

services during lockdown by the
COVID-19 survey in five UK national
longitudinal studies
Constantin-­Cristian Topriceanu ‍ ‍,1,2 Andrew Wong,2 James C Moon,3,4
Alun D Hughes,1,2 David Bann,5 Nishi Chaturvedi,1,2 Praveetha Patalay,2,5
Gabriella Conti,6 Gaby Captur1,2,7

To cite: Topriceanu C-­C, ABSTRACT

Wong A, Moon JC, et al. Strengths and limitations of this study
Objective Access to health services and adequate care
Evaluating access to health
is influenced by sex, ethnicity, socioeconomic position ►► A strength of the study was the implicit age homo-
and care services during
lockdown by the COVID-19
(SEP) and the burden of comorbidities. Our study aimed to geneity of birth cohort participants. This enabled age
survey in five UK national assess whether the COVID-19 pandemic further deepened matching within each cohort for each data collection
longitudinal studies. BMJ Open these already existing health inequalities. sweep.
2021;11:e045813. doi:10.1136/ Design Cross-­sectional study. ►► Combining five cohorts spanning multiple age
bmjopen-2020-045813 Setting Data were collected from five longitudinal age-­ groups (19, 30, 50, 62 and 74 years old) led to a bet-
►► Prepublication history and
homogenous British cohorts (born in 2000-2002, 1989- ter understanding of how the COVID-19 pandemic
additional material for this 1990, 1970, 1958 and 1946). has affected different generations.
paper are available online. To Participants A web survey was sent to the cohorts. ►► The longitudinal nature of the cohorts allowed the
view these files, please visit Anybody who responded to the survey was included, derivation of individualised non-­response weights
the journal online (http://​dx.​doi.​ resulting in 14 891 eligible participants. for each of the included participants. This enabled us
org/​10.​1136/​bmjopen-​2020-​ Main outcomes measured The survey provided data to address non-­response bias, rendering our results
045813).
on cancelled surgical or medical appointments, and the more generalisable.
Received 13 October 2020 number of care hours received in a week during the first ►► As self-­reported measures were used, the number
Revised 04 February 2021 UK COVID-19 national lockdown. of care hours needed were subjected to report-
Accepted 01 March 2021 Interventions Using binary or ordered logistic regression, ing biases. In addition, single categorical outcome
we evaluated whether these outcomes differed by sex, variables do not have the capacity to measure the
ethnicity, SEP and having a chronic illness. Adjustment impact spectrum generated by the cancelled ap-
was made for study design, non-­response weights, pointments or to fully capture the care needs of the
psychological distress, presence of children or adolescents participants.
in the household, COVID-19 infection, key worker status, ►► We binarised the ethnicity variable to enable suffi-
and whether participants had received a shielding letter. cient sample sizes for comparisons, but this preclud-
ed more detailed comparisons between the diverse
Meta-­analyses were performed across the cohorts, and
ethnic groups which exist in the UK. Older cohorts
meta-­regression was used to evaluate the effect of age as
(50, 62 and 74 years old) consisted of almost only
a moderator.
white participants, so we were unable to describe
Results Women (OR 1.40, 95% CI 1.27 to 1.55) and
findings for older people from ethnic minority groups
those with a chronic illness (OR 1.84, 95% CI 1.65 to
who might have been most adversely affected by
2.05) experienced significantly more cancellations during the lockdown. Our younger cohorts (19 and 30 years
lockdown (all p<0.0001). Ethnic minorities and those with old) included less than 20% non-­white participants,
a chronic illness required a higher number of care hours which could result in specious associations. Our
during the lockdown (both OR≈2.00, all p<0.002). SEP was analysis did not take into account racism as a struc-
not associated with cancellation or care hours. Age was tural actor to explain the disparities observed, and
not independently associated with either outcome in the further work will be needed to address this.
© Author(s) (or their
meta-­regression.
employer(s)) 2021. Re-­use
permitted under CC BY. Conclusion The UK government’s lockdown approach
Published by BMJ. during the COVID-19 pandemic appears to have deepened INTRODUCTION
For numbered affiliations see existing health inequalities, impacting predominantly On 11 March 2020, the WHO declared the
end of article. women, ethnic minorities and those with chronic illnesses.
novel SARS-­CoV-2 (also known as COVID-19)
Public health authorities need to implement urgent policies
Correspondence to outbreak a global pandemic. As the UK was
to ensure equitable access to health and care for all in
Dr Gaby Captur; facing a surge of new cases, the government
preparation for a fourthwave.
​gabriella.​captur@​ucl.​ac.​uk imposed national lockdown restrictions on

Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813 1

Open access

23 March 2020 in order to limit the spread of the virus. the end of the survey, participants also had the option
Although the restrictions were gradually relaxed, the most to enter free text to describe their particular COVID-19
widely accepted end date of the lockdown was consid- experience. There were no inceptives, just an invite to
ered to be on 4 July 2020, when non-­essential businesses participate sent by email. Two email reminders were sent
such as bars and restaurants opened. Delivery of routine to NCDS, BCS70, NS and MCS participants who had
care across the UK National Health Service (NHS) was not started or who had partially completed the survey.
hampered by the pandemic crisis and the lockdown. A single email reminder was sent to NSHD participants.
Access to health services and adequate care has been
previously shown to be influenced by sex, ethnicity, socio-
economic position (SEP) and the burden of comorbidi- Patient and public involvement
ties.1 2 However, it is unknown whether access to health It was not appropriate or possible to involve the study
and care services during the COVID-19 pandemic participants or the public in the design or conduct of our
differed by these factors, potentially further widening research. However, we plan to disseminate the results to
already existing health inequalities.3 Evidence from the study participants.
previous pandemics suggests this possibility, but data are
missing in the context of COVID-19 currently. To answer
Outcomes
these questions, a web-­based survey was sent to partici-
Individuals experiencing a healthcare-­ related cancella-
pants in five UK national longitudinal studies, spanning
tion in the form of a cancelled surgery, medical proce-
multiple generations; data were collected during the
dure or other medical appointment at any time since the
core UK lockdown, between 2 May and 1 June 2020. The
beginning of the SARS-­CoV-2 pandemic were scored as 1,
survey questions can be accessed online (https://​cls.​ucl.​
or 0 otherwise. The number of hours of help received for
ac.​uk/​wp-​content/​uploads/​2020/​12/​COVID-​19-​Online-​
self or other household member in a typical week during
Survey-​Questionnaire-​Wave-​1-​April-​2020-​Version-​2.​pdf).
lockdown was recorded in six categories: 0, 1–4, 5–9,
We investigated the number of participants who had
10–19, 20–34 and 35+ hours. The variable encompasses
a cancelled surgical or medical appointment, and the
both home healthcare and social help. The care provider
number of care hours received for self or other household
could be a family member, a friend, a professional paid
members over a week during the lockdown. We analysed
carer or a voluntary helper. Thus, we had two outcomes:
how these outcomes varied by already established factors
cancelled surgery, medical procedures or other medical
contributing to health inequalities. The importance of
appointments, and the number of hours of help received
cancellations stems from the potential consequences of
healthcare deprivation, while the number of weekly care in a week.
hours has been shown to predict admission to long-­term
care facilities especially in the older population.4 Exposures
Sex was recoded as 0=male and 1=female, while ethnicity
was recoded as 0=non-­ white and 1=white. As NSHD,
METHODS NCDS and BCS70 consisted mostly of white participants,
Study design ethnicity data were examined only for the NS and MCS
The five UK national longitudinal studies were the cohorts. Highest educational attainment and financial
National Study of Health and Development (NSHD),5 the difficulties prior to COVID-19 were used as a proxy for
National Child Development Study (NCDS),6 the 1970 adult SEP. Highest educational attainment was catego-
British Cohort Study (BCS70),7 Next Steps (NS)8 and the rised as degree/higher, advanced-­ level exam/diploma,
Millennium Cohort Study (MCS).9 NSHD participants ordinary-­ level exam/general certificate of secondary
were born in 1946, NCDC in 1958, BCS70 in 1970 and education or none. Financial difficulties before lockdown
MCS in 2000–2002, and all participants were followed up were self-­ rated using the following options: managing
from birth (all birth cohorts), while NS is a longitudinal comfortably, all right, getting by and difficult. As many
cohort study whose participants, born in 1989–1990, were MCS participants were still undertaking education
followed up from adolescence. The cohorts were exten- and were financially dependent on their families, their
sively followed up with periodic assessments which have parents’ highest education and financial difficulties were
been described elsewhere. During the UK COVID-19 May used. Childhood social class was recorded according
2020 lockdown, an online questionnaire was sent to each to the UK Office of Population Censuses and Surveys
participant from each cohort. The questionnaire was Registrar General’s social class, resulting in six cate-
designed to explore the physical health, health behaviours, gories: professional, managerial and technical, skilled
social contact and support, loneliness and mental health, non-­manual, skilled manual, partly skilled or unskilled.
household relationships and care needs, housing situa- Participants were asked to report whether they had a
tion, employment, finances and benefits and education long-­standing illness (yes/no). In addition, the nature of
during the height of the COVID-19 pandemic. The ques- each chronic illness was also broadly recorded. Thus, our
tionnaire's format was mostly multiple choice with either exposures were sex, ethnicity, SEP and the presence of a
binary (yes/no) or categorical response options. Towards chronic illness.

2 Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813

 Open access

Covariates appointments, while ordinal logistic regression was

Participants who were clinically extremely vulnerable and used to predict the number of care hours received. The
at high risk of complications from a potential SARS-­CoV-2 proportional odds assumption for ordinal logistic regres-
infection were sent a letter or text message from the NHS sion was tested using a Brant test.18 Weights to account
or the chief medical officer advising them to shield, that for the stratified survey designs of 1946, 1990 and 2000–
is, to stay at home except for specific purposes and to 2002 cohorts19 have been previously developed. Logistic
avoid contact with people they do not live with, except regression models predicting each participant’s response
for specific purposes. Receipt of such a shielding letter during the COVID-19 data sweep based on known demo-
was recorded as a binary variable (yes/no). The number graphic, socioeconomic, household and individual
of hours of help received for self or another household predictors of non-­response at previous data collection
member in a week before the pandemic was recorded as points were used to calculate non-­response weights.20 21
mentioned earlier. In the logistic regression models, missing covariate values
Women were more likely to care for children aged less were generated using multiple imputation. For each
than 16 years and were more likely to report psycholog- COVID-19 survey respondent, the probability of response
ical distress.10 Thus, sex differences were further explored was calculated, and non-­response weights were derived as
using the two covariates. The presence of children aged the inverse probability of response with further calibra-
less than 16 years in the household and the self-­reported tion to sum to the total number of respondents in each
presence of psychological distress during lockdown were cohort. The stratified survey design and non-­response
recorded (yes/no). Psychological distress was measured weights were combined to generate a combined weight.16
using the 12-­ Item General Health Questionnaire,11 An individualised combined weight was derived for each
defined as 0 if <4, and one if ≥4 for NSHD and NS, and study respondent (full details available in the Centre
using a shortened 9-­item Malaise Inventory defined as 0 for Longitudinal Study COVID-19 Survey User Guide,
if <4, and one if ≥4 for NCDS and BCS70 as previously https://​ cls.​ u cl.​ a c.​ u k/​ w p-​ content/​ u ploads/​ 2020/ ​12/​
described.12 For MCS, the Kessler K6 score was used, UCL-​Cohorts-​COVID-​19-​Survey-​user-​guide.​pdf). Predic-
defined as 1 if ≥13 and 0 otherwise.13 tors were included sequentially one at a time. Sex analyses
In the literature, non-­ white ethnicity was associated were adjusted for these individualised combined weights
with key-­worker status14 and COVD-19 infection.15 Thus, and for the receipt of a shielding letter. All other analyses
ethnicity differences were further explored using the were similarly adjusted, but ethnicity analyses were addi-
two covariates. Key worker status was self-­rated based on tionally adjusted for sex; SEP analyses were additionally
whether the participants’ work was classified as critical to adjusted for sex and ethnicity; and chronic illness anal-
the COVID-19 response. COVID-19 infection was recoded yses were additionally adjusted for sex, ethnicity and SEP.
as 0=no and 1=yes, based on a positive antigen or anti- Gender differences were further evaluated by adjusting
body test or strong personal suspicion due to symptoms. for the presence of children aged less than 16 years in the
household and for psychological distress during the lock-
Statistical analysis down. As women are more likely to have a chronic disease,
Statistical analysis was performed in R V.3.6.0. Frequency gender differences were also evaluated after adjustment
distribution of continuous data was assessed visually using for the presence of a chronic disease.22 Ethnicity differ-
histograms. Categorical variables were expressed as counts ences were further explored by adjusting for key worker
and percentages for each available category. Within each status as ethnic minorities have been reported to be
cohort, childhood SEP, highest educational attainment over-­represented as key workers in the literature.14 As
and financial difficulties were converted into cumulative other studies have shown greater COVID-19 positivity
rank probabilities (ridit scores) to quantify the difference rates among ethnic minorities,15 further adjustment
in outcomes comparing the lowest with highest SEP (ie, for COVID-19 infection was pursued for ethnicity in all
the relative indices of inequality).16 Models containing analyses.
all socioeconomic variables were assessed for multicol- We also explored whether individuals with multiple
linearity via the variance inflation method. As child- comorbidities (defined as two or more) were more likely
hood SEP was multicollinear with the other two, had the to experience a cancelled appointment or require a
least amount of missing data and could impact on adult higher number of care hours compared with individuals
behaviours and health outcomes independently of adult with a single long-­ standing illness. Comparisons were
SEP,17 it was used in the subsequent analysis. However, made using χ2 test for the cancelled appointment anal-
we additionally report the results from the analyses using yses and Mann-­Whitney U test for the care hours analyses.
SEP based on highest educational attainment and finan- Cohort-­specific analyses were conducted initially.
cial difficulties, respectively. Meta-­analyses were then performed across the cohorts,
Separate regression models were employed using sex, only if there was a significant result in at least one of
ethnicity, SEP and presence of chronic illness as predic- the cohorts. Heterogeneity was evaluated using the
tors of cancelled appointments or number of care hours Cochran Q test and Higgins I2 statistic. As smaller
received during lockdown. Generalised linear models samples have more sampling errors in their effect esti-
with logit link were employed to predict cancelled mate, larger effect size might emerge.23 Thus, funnel

Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813 3

Open access

plot asymmetry was evaluated using the Egger test. RESULTS

Metaregression was conducted with age/cohort as Overall, 15 291 participants (45% of the combined
a moderator in order to determine whether it was a cohorts’ participants) responded to the COVID-19 survey
source of heterogeneity. As the associations between as follows: 1241 out of 1842 (NSHD), 5205 out of 8943
age and our outcomes were likely to be non-­ linear (NCDS), 4247 out of 10 458 (BCS), 1921 out of 9380
based on visual inspection, we performed the metare- (NS) and 2677 out of 9909 (MCS). Being female, having
gression using restricted cubic splines modelling.24 a higher educational attainment, having a higher income
We ran sensitivity analyses in which we (1) simulated a and reporting better self-­ rated health were associated
complete case analysis through multiple imputation to with higher response rates.19
verify the reliability of the observed sex-­related differ-
Any participant who lacked data for at least one
ences as the majority of our respondents were female;
outcome variable was excluded, leaving 14 891 eligible
using the predictive mean matching method, we have
participants who were included in the final analysis
generated five complete data sets25 and performed
a pooled regression; the models were not further (characteristics summarised in table 1). A breakdown
adjusted for non-­response weights; (2) adjusted the of data missingness is presented in online supple-
number of care hours received during lockdown anal- mental table S1. Overall, included participants were
yses for the number of care hours received before the more likely to be female, over 50 years of age and of
pandemic; (3) explored possible deviations from the a higher educational attainment. Older participants
proportional odds assumption via multinomial logistic were more likely to have a chronic illness, receive a
regression with the number of care hours grouped shielding letter, experience a cancelled appointment
into never (0 hours), low (1–9 hours) and high (10+ and require more care hours during lockdown. The
hours). chronic illnesses reported spanned a variety of medical

Table 1 Characteristics of participants by cohort

Cohort study birth year
Participant characteristics 1946 1958 1970 1989–1990 2000–2002
Sample size
 Questionnaire respondents (n=15 291) 1241 5205 4247 1921 2677
 Included participants (n=14 891) 1154 5119 4131 1876 2609
Age (years) 74 62 50 30–31 19–20
Male (%) 607 (51.88) 2432 (47.51) 1708 (41.40) 633 (34.09) 770 (29.51)
Non-­white ethnicity (%) N/A N/A N/A 361 (19.27) 367 (14.17)
Childhood SEP I–III (%) 633 (57.18) 1897 (43.60) 1727 (48.08) 1227 (69.36) 1755 (79.70)
Chronic Illness (%) 842 (73.02) 3099 (61.24) 1955 (48.08) 715 (39.20) 830 (33.33)
Multimorbidity (%) 390 (33.33) 1408 (27.83) 739 (18.18) 194 (10.64) 165 (6.63)
Shielding letter (%) 112 (9.61) 334 (6.57) 196 (4.77) 56 (3.00) 60 (2.30)
Presence of children<16 years (%) 0 (0.00) 87 (2.13) 1660 (41.10) 462 (25.37) 15 (0.60)
Psychological distress during lockdown (%) 216 (18.77) 452 (10.25) 556 (16.07) 655 (39.15) 188 (8.29)
Key workers (%) 9 (0.78) 938 (18.32) 1396 (33.79) 583 (31.08) 196 (7.51)
COVID-19 infection-­self-­reported or positive test (%) 27 (2.31) 296 (5.78) 379 (9.18) 197 (10.50) 158 (6.06)
COVID-19 infection-­positive test only (%) 1 (0.09) 19 (0.37) 17 (0.41) 12 (0.64) 7 (0.27)
Outcomes
Cancelled appointments (%) 376 (32.58) 775 (15.17) 494 (11.97) 234 (12.47) 303 (11.61)
Care hours during lockdown
 0 1073 4651 3825 1724 2552
 1–4 61 112 66 47 47
 5–9 10 41 36 10 4
 10–19 8 42 20 17 3
 20–34 5 18 16 4 2
 35+ 13 54 35 10 1

1946 refers to NSHD; 1958 refers to NCDS; 1970 refers to BCS70; 1989–1990 refers to NS; 2000–2002 refers to MCS.
BCS70, 1970 British Cohort Study; MCS, Millennium Cohort Study; N/A, not available; NCDS, National Child Development Study; NS, Next Steps;
NSHD, National Study of Health and Development; SEP, socioeconomic position.

4 Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813

 Open access

Table 2 Association of sex, ethnicity, SEP and the presence of chronic illness with cancelled surgery, medical procedures or
other medical appointments during lockdown
Sex* Ethnicity† SEP‡ Chronic illness§
Cohort study
birth year OR (95% CI) P value OR (95% CI) P value OR (95% CI) P value OR (95% CI) P value
1946 0.97 (0.76 to 1.25) 0.827 N/A N/A 1.39 (0.90 to 2.16) 0.138 1.74 (1.28 to 2.36) 0.0004
(n=1170)
1958 1.20 (1.03 to 1.40) 0.021 N/A N/A 1.05 (0.78 to 1.41) 0.753 2.15 (1.76 to 2.62) <0.0001
(n=5073)
1970 1.83 (1.47 to 2.26) <0.0001 N/A N/A 1.05 (0.73 to 1.51) 0.786 1.77 (1.42 to 2.21) <0.0001
(n=4099)
1989–1990 1.70 (1.23 to 2.35) 0.001 1.25 (0.86 to 2.37) 0.255 1.45 (0.88 to 2.41) 0.154 1.59 (1.18 to 2.13) 0.002
(n=1849)
2000–2002 2.29 (1.65 to 3.19) <0.0001 1.03 (0.73 to 2.31) 0.885 1.05 (0.66 to 1.67) 0.836 1.71 (1.30 to 2.25) 0.0001
(n=2605)

All analyses used generalised linear models with logit link. Significant p values are highlighted in bold.
*Sex was coded as 0=male and 1=female; adjustment was made for survey combined weight and shielding letter.
†Ethnicity was coded as 0=non-­white and 1=white; adjustment was made for survey combined weight, shielding letter and sex. Almost all
participants in NSHD (1946), NCDS (1958) and BCS (1970) were white, so ethnicity was not examined.
‡SEP was coded using childhood social class from 1=managerial to 6=unskilled, but ridit scores were used in all analyses; adjustment was made for
survey combined weight, shielding letter, sex and ethnicity.
§Chronic illness was coded as 0=absent and 1=present; adjustment was made for survey combined weight, shielding letter, sex, ethnicity and SEP.
N/A, not available; NCDS, National Child Development Study; NSHD, National Study of Health and Development; SEP, socioeconomic position.

systems (online supplemental table S2). Across all during lockdown (online supplemental table S5) attenu-
cohorts, the most prevalent conditions were high ated the regression coefficients in most cohorts, but sex
blood pressure (2119 participants), recurrent back differences persisted. All the sex differences persisted
problems (1884 participants), mental health issues after adjusting for the presence of a chronic illness, but
(1708 individuals) and asthma (1703 individuals). most coefficients were attenuated (online supplemental
Individuals with multiple comorbidities were more table S6). The meta-­analysis revealed a pooled OR of
likely to experience cancelled surgeries, medical
1.40 (95% CI 1.27 to 1.55) in the absence of funnel plot
procedures or medical appointments during lock-
asymmetry (Egger test, p=0.376; table 3). However, there
down and to require more care hours than those with
a single chronic condition (online supplemental table was considerable heterogeneity between the cohorts
S3). In NS, non-­white participants were less likely to (I2=85.78%, p<0.0001). In each of the cohorts and in the
be key workers (p=0.016), but we found no association meta-­analysis, the presence of a chronic illness at baseline
between ethnicity and COVID-19 infections (p=0.296). was associated with higher odds (pooled OR 1.84, 95% CI
In MCS, there was no association between ethnicity, 1.65 to 2.05) of experiencing a cancelled event. The meta-­
and neither being a key worker (p=0.647) nor being analysis revealed no heterogeneity (I2=0.00%, p=0.422)
infected with COVID-19 (p=0.979). and there was no evidence of funnel plot asymmetry
when using the SE as the predictor (Egger test p=0.092).
Cancelled surgery, medical procedures or medical
Ethnicity and SEP were not associated with cancella-
appointments during lockdown
tions in any of the cohorts. Age was not significant in the
In all cohorts except NSHD, female sex was associated
with higher odds (OR range 1.20–2.29, all p<0.021) metaregression (online supplemental table S7). A visual
of cancelled surgery, medical procedures or medical representation of the cancelled surgery, medical proce-
appointments (table 2). Adjusting for the presence of dures or medical appointments by sex, ethnicity and the
children less than 16 years old (online supplemental presence of chronic illness across the five UK cohorts is
table S4) and for the presence of psychological distress presented in figure 1.

Table 3 Meta-­analysis for the respective association of sex and presence of chronic illness with cancelled surgery, medical
procedures or other medical appointments during lockdown
Study heterogeneity Egger test
Predictor N I2 Q P value OR (95% CI) P value P value
Sex 14 796 85.78% 28.12 <0.0001 1.40 (1.27 to 1.55) <0.0001 0.376
Chronic illness 12 584 0.00% 3.89 0.422 1.84 (1.65 to 2.05) <0.0001 0.092
Significant p values are highlighted in bold.

Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813 5

Open access

Figure 1 Bar charts illustrating the percentages of cancelled surgery, medical appointments or other medical procedures
by sex, ethnicity and the presence of chronic illness across the five UK longitudinal cohorts, ordered by increasing age of the
cohort from left to right. Error bars representing the 95% CIs are also presented. 1946 refers to NSHD; 1958 refers to NCDS;
1970 refers to BCS70; 1989–1990 refers to NS; 2000–2002 refers to MCS. BCS70, 1970 British Cohort Study; MCS, Millennium
Cohort Study; NCDS, National Child Development Study; NS, Next Steps; NSHD, National Study of Health and Development.

Number of care hours for self or another household member to low (1–9 hours), but more variability was observed at
during lockdown the transition from low to high (10+ hours, online supple-
In older cohorts, chronic illness was more prevalent, and mental table S12). When using the highest educational
the association with number of care hours needed was attainment or financial difficulties as socio-­economical
stronger (table 4). In the meta-­analysis, a higher number surrogates instead of childhood SEP, there were still no
of care hours was associated with ethnic minorities (OR significant associations with cancellations (online supple-
0.53, 95% CI 0.35 to 0.79, I2=34.17%) and with the pres- mental table S13) or the number of care hours during
ence of chronic illness (OR 2.20, 95% CI 1.72 to 2.56, lockdown (online supplemental table S14).
I2=13·22%; table 5). After adjusting for key worker status,
significant associations persisted (online supplemental
table S8). After adjusting for COVID-19 infection, signif- DISCUSSION
icant associations persisted (online supplemental table Statement of principal findings
S9). Sex and SEP were not associated with the number These data from five UK national longitudinal studies,
of care hours received during lockdown. There was no at the height of the UK national COVID-19 lockdown in
evidence that age contributed to the heterogeneity May 2020, indicate worrying health inequalities in the
between cohorts from the metaregression (online supple- access to health and care services—worst hit were women
mental table S7). Visual representation of the data is and those with a chronic illness or from ethnic minority
provided in figure 2. groups.

Sensitivity analysis Meaning of the study

Associations between sex and cancelled surgery, medical Even before the COVID-19 lockdown, persons with
procedures or other medical appointments persisted chronic illnesses were vulnerable26 27 and required more
after multiple imputation (online supplemental table access to health services, as well as care from family
S10). Adjustment for thr number of care hours received members, friends and care service providers.28 The
before the pandemic attenuated the ORs, but the associ- pandemic triggered unprecedented changes affecting
ations mostly persisted (online supplemental table S11). healthcare (which shifted to prioritise COVID-19
Findings were similar in the multinomial logistic regres- patients) and socioeconomic dynamics (caused by
sion when looking at the transition from never (0 hours) restricted movement, changes to work patterns and

6 Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813

 Open access

remuneration, and unstable housing). Our results show

Brant test
that participants with chronic illnesses were twice as

P value

‡SEP was coded using childhood social class from 1=managerial to 6=unskilled, but ridit scores were used in all analyses; adjustment was made for survey combined weight, shielding letter, sex and
†Ethnicity was coded as 0=non-­white and 1=white; adjustment was made for survey combined weight, shielding letter and sex. Almost all participants in NSHD (1946), NCDS (1958) and BCS (1970)
likely to have cancelled medical appointments, poten-

0.192
0.860

0.244

0.010

0.961
tially depriving them of vital medical care. They were
also twice as likely to require increased number of care
<0.0001

<0.0001
P value
0.009

0.047

0.301
hours. Only around 50% of the participants had their
care hours expectations met, which suggests that a
significant proportion were deprived of essential care.
2.20 (1.22 to 3.99)
2.17 (1.56 to 3.04)

2.74 (1.84 to 4.08)

1.63 (1.01 to 2.65)

1.38 (0.75 to 2.56)

Results persisted after adjustment for shielding letter
Chronic illness§

and previous care hours, illustrating their deeply rooted

OR (95% CI)

associations with the outcomes. Overall, participants

with chronic illnesses received a double hit with poten-
tially long-­lasting effects on their health and well-­being.
These negative effects were even more pronounced for
Brant test

individuals suffering from multiple comorbidities.

P value

We found that women were more likely to experience

0.0003
0.329
0.877

0.026

0.998

§Chronic illness was coded as 0=absent and 1=present; adjustment was made for survey combined weight, shielding letter, sex, ethnicity and SEP.
cancellations in planned surgery, medical procedures
Association of sex, ethnicity, SEP and the presence of chronic illness with number of care hours during lockdown

or other medical appointments during lockdown. This

P value

could be linked to pre-­existing sex inequalities where

0.683
0.282

0.272

0.727

0.146

women adopt a more caring role prioritising other

family members’ needs over their own.29 Sex inequalities
1.17 (0.56 to 2.45)
1.29 (0.81 to 2.06)

0.72 (0.39 to 1.30)

0.87 (0.39 to 1.94)

2.17 (0.77 to 6.25)

during lockdown could also have widened on account

of the added childcare responsibilities, including home
OR (95% CI)

schooling, being predominantly undertaken by women.

Adjusting for the presence of children under 16 years
SEP‡

in the household attenuated the regression coefficients,

*Sex was coded as 0=male and 1=female; adjustment was made for survey combined weight and shielding letter.

suggesting this was a likely contributory factor.

All analyses used generalised linear models with ordinal logit link. Significant p values are highlighted in bold.

Ethnic minorities were twice as likely to require an

Brant test
P value

increased number of care hours compared with white

0.005

0.972

participants in the younger cohorts. It is likely that the

N/A
N/A

N/A

unstable socioeconomic landscape dominated by loss

P value

of income, unstable housing, increased psychological

0.001

0.432

distress and reduced community support brought about

N/A
N/A

N/A

by the lockdown restrictions adversely impacted these

0.44 (0.27 to 0.72)

0.76 (0.38 to 1.53)

communities. Our care hours variable captures both

home healthcare and social needs, potentially high-
OR (95% CI)

lighting broad extra needs during lockdown. Another

Ethnicity†

explanation could stem from the fact that ethnic minori-

ties are over-­represented as key workers.14 To meet the
N/A
N/A

N/A

care needs of their communities, they could have been

subjected to increased working hours, unusual working
Brant test

<0.0001

<0.0001
P value

environments, stricter work-­based controls and greater

0.656

0.010

0.998

exposure to COVID-19, exacerbating both physical and

psychological stress. However, our data suggest that
N/A, not available; SEP, socioeconomic position.
P value

ethnic minorities were under-­represented as key workers

0.452
0.087

0.955

0.102

0.681

in the younger cohorts (NS and MCS).

were white, so ethnicity was not examined.

Although the NHS has an extensive coverage and is

free at the point of use, healthcare inequalities have been
1946 (n=1170) 1.17 (0.77 to 1.79)
1.25 (0.70 to 1.61)

0.99 (0.72 to 1.36)

0.69 (0.44 to 1.08)

0.88 (0.49 to 1.64)

reported in the UK in the past decade.30 An important

OR (95% CI)

negative finding was the absence of an association

between lower SEP and access to health and care services
during lockdown. Speculatively, this could mean that the
Sex*

multiple policies implemented by the UK government to

address such inequalities have paid off.
Cohort study

Rather surprisingly, the meta-­regression showed that

1989–1990

2000–2002
birth year

(n=4884)

(n=3972)

(n=1787)

(n=2605)
Table 4

ethnicity.

age was not a predictor for cancellations or accentuated

1958

1970

care needs, suggesting an age-­ homogenous effect of

the lockdown across the generations. We expected that

Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813 7

Open access

Table 5 Meta-­analysis for the respective association of sex and presence of chronic illness with number of care hours during
lockdown
Study heterogeneity
Predictor N I2 Q P value OR (95% CI) P value Egger test P value
Ethnicity 4371 34.17% 0.218 0.218 0.53 (0.35 to 0.79) 0.002 N/A*
Chronic illness 12 684 13.22% 4.609 0.330 2.10 (1.72 to 2.56) <0.0001 0.312
Significant p values are highlighted in bold.
*Egger test was not feasible as only two studies recorded ethnicity.
N/A, not available.

older generations, being more frail and likely to have system. The challenge facing public health authorities
received a shielding letter than younger persons, would is the need to promote access to healthcare for vulner-
have required more care for activities of daily living able groups on the one hand while minimising infec-
and clinical appointments during lockdown. However, tion exposure on the other. Countries without a free
the younger generations (NS and MCS) were similarly healthcare system where citizens rely on paid insur-
affected in terms of medical appointment cancellations, ance, such as the USA, are in an even more difficult
as well as the number of care hours required during position.33
the pandemic. This is a potentially worrisome indica-
tion that the disruption caused by lockdown may have Unanswered questions and future research
had far-­reaching effects on the health and well-­being of
Remote healthcare known as telehealth, has been
young people in the UK.
brought forward as a potential solution to the
Implications for clinicians and policymakers problem of health inequalities in the COVID-19 situ-
As pandemics can be characterised by multiple waves, ation. However, telehealth is fraught with similar
they can last several years.31 Given the prospect of a digital inequalities that will hamper the provision of
fourth wave, it is vital that public health authorities equitable access.34 35 To make telehealth egalitarian,
implement national interventions to bolster health factors contributing to digital inequalities need to be
and care access. In addition, the healthcare disrup- addressed. These include technical hardware dispar-
tions that occurred during the first wave are expected ities (lack of technological equipment and slower
to lead to a surge in late-­presenting conditions such internet connections), digital literacy and access to
as cancer,32 which will further strain the healthcare technical support.

Figure 2 Bar charts illustrating the percentage of participants requiring support based on the number of care hours needed
during the UK COVID-19 national lockdown stratified by sex, ethnicity and the presence of chronic illness across the cohorts.

8 Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813

 Open access

Strengths and weaknesses of the study number of knots per variable as our study included only
Strengths of the study are the implicit age homoge- five cohorts.36
neity of participants, enabling age matching within
each cohort as participants were exposed to similar life Strength and weakness in relation to other studies
factors before the national lockdown. Combining five To the best of our knowledge, this is the first UK study
cohorts spanning multiple age groups (19, 30, 50, 62 and to highlight worrying health inequalities in the access
74 years old) led to a better understanding of how the to health and care services as a result of the COVID-19
COVID-19 pandemic has affected different generations. pandemic. In addition, we are the first to address this
The longitudinal nature of the cohorts allowed the deri- issue by combining multiple cohorts spanning multiple
vation of individualised non-­response weights for each age groups and with such a high sample size.
of the participants,20 21 which have been included in all
analyses. This enabled us to address non-­response bias,
rendering our results more generalisable. CONCLUSION
Limitations include data missingness due to low Individuals with a chronic illness were more likely to expe-
response rates, particularly in younger cohorts, and rience cancelled healthcare appointments and greater
small sample sizes of older cohorts, particularly NSHD. care needs during the UK national lockdown generated by
As self-­reported measures were used, the number of the COVID-19 pandemic. Women experienced reduced
care hours needed before and during lockdown were access to healthcare, while ethnic minorities required
subject to reporting biases. We binarised the ethnicity extra care hours. Our results suggest that the pandemic
variable to enable sufficient sample sizes for compar- might have widened pre-­existing healthcare inequalities,
isons, but this precluded more detailed comparisons further depriving already vulnerable and disadvantaged
between the diverse ethnic groups which exist in the groups of the health and care services which they need.
UK. Older cohorts (NSHD, NCDS and BCS70) consist Public health measures should be rapidly implemented
of almost only white participants, so we were unable to to better protect and meet the health and care demands
describe findings for older people from ethnic minority of such at-­risk groups ahead of a COVID-19 third wave.
groups that may have been most adversely affected by
lockdown. Our younger cohorts (NS and MCS) include Author affiliations
1
School of Medicine, University College London, London, UK
less than 20% non-­white participants, which could result 2
UCL MRC Unit for Lifelong Health and Ageing, University College London, London,
in specious associations. Our analysis did not take into UK
account racism as a structural actor to explain the dispar- 3
Institute of Cardiovascular Science, University College London, London, UK
ities observed, and further work will be needed in future 4
Cardiac Imaging Department, Barts Heart Center, London, UK
5
studies to address this. By considering chronic illness as a Center for Longitudinal Studies, Department of Social Science, University College
London, London, UK
binary variable, we were unable to discriminate between 6
Department of Economics and UCL Social Research Institute, University College
minor and serious illnesses, capture multimorbidity or London, London, UK
measure the impact spectrum generated by the cancelled 7
Center for Inherited Heart Muscle Conditions, Cardiology Department, The Royal
appointments as well as the loss of care hours. In addi- Free Hospital, London, UK
tion, we have not collected any data on the severity of the
chronic diseases which could directly influence the need Twitter Constantin-­Cristian Topriceanu @CTopriceanu
of medical appointments as well as the number of care Acknowledgements The authors thank all the members of our studies for their
hours required in a week. We did not capture whether contribution to this COVID-19 survey and for their ongoing participation in our
studies. We thank the survey, data and administrative teams at the Center for
participants had more than one cancelled appointment Longitudinal Studies and Unit for Lifelong Health and Ageing, UCL, for enabling the
or procedure. This is especially relevant for participants rapid COVID-19 data collection to take place.
from lower socioeconomic backgrounds who are more Contributors CCT analysed the data and wrote the manuscript. CCT, JCM, ADH,
likely to have multiple comorbidities and may have had DB, NC, PP, GCo and GCa were involved in the study design and implementation.
more than one cancelled appointment or procedure. By AW, ADH, DB, NC, PP and GCo actively participated in the data acquisition. GCa
contributed to the data analysis, interpretation of the results and manuscript
reducing many regression variables to a binary coding,
drafting; was the guarantor of this work; and attested that all listed authors met the
we may have underestimated socioeconomic differences authorship criteria and that no others meeting the criteria were omitted. All authors
which were recorded over a wider categorical spectrum. were involved in critically revising the manuscript and approving the final version.
The survey question about cancelled appointments did Funding The study was funded by the Economic and Social Research Council
not distinguish between face-­to-­face and virtual clinic under the Center for Longitudinal Studies, Resource Center 2015–2020 (grant
consultations. We were unable to separate pandemic number ES/M001660/1) and by the Medical Research Council (grant MC_
UU_00019/1). GCa is supported by British Heart Foundation (MyoFit46 Special
effects from recognised confounders such as seasonal Programme Grant SP/20/2/34841), the National Institute for Health Research Rare
variation in the number of care hours needed, as well Diseases Translational Research Collaboration (NIHR RD-­TRC) and by the NIHR UCL
as other unobserved confounders. The overall preva- Hospitals Biomedical Research Center. JCM is directly and indirectly supported
lence of outcomes differed between cohorts, and this by the UCL Hospitals NIHR BRC and Biomedical Research Unit at Barts Hospital,
respectively. DB is supported by the Economic and Social Research Council (grant
can affect the interpretation of ORs, potentially intro- number ES/M001660/1) and by The Academy of Medical Sciences/Wellcome Trust
ducing bias between cohort comparisons. Lastly, a limita- (Springboard Health of the Public in 2040 award: HOP001/1025). AH receives
tion of the restricted cubic spline metaregression is the support from the British Heart Foundation, the Economic and Social Research

Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813 9

 Open access

Council, the Horizon 2020 Framework Programme of the European Union, the British National child development study (NCDS). BMC Psychol
National Institute on Aging, the National Institute for Health Research University 2016;4:2016.
College London Hospitals Biomedical Research Center, and the UK Medical 13 Stolk Y, Kaplan I, Szwarc J. Clinical use of the Kessler psychological
Research Council and works in a unit that receives support from the UK Medical distress scales with culturally diverse groups. Int J Methods Psychiatr
Res 2014;23:161–83.
Research Council. GCo thanks for support the European Research Council under 14 CDC. Health equity considerations and racial and ethnic
the European Union’s Horizon 2020 research and innovation programme (grant minority groups: centers for disease control and prevention,
agreement number 819752 DEVORHBIOSHIP–ERC-2018-­COG). 2020. Available: https://www.​cdc.​gov/​coronavirus/​2019-​ncov/​
Competing interests None declared. community/​health-​equity/​race-​ethnicity.​html [Accessed 21 Aug
2020].
Patient consent for publication Not required. 15 Sze S, Pan D, Nevill CR, et al. Ethnicity and clinical outcomes in
COVID-19: a systematic review and meta-­analysis. EClinicalMedicine
Ethics approval Ethical approval was obtained from relevant committees and from 2020;29-30:100630-30:29–30.
the University College London/Institute of Education research ethics committee 16 Bann D, Johnson W, Li L, et al. Socioeconomic inequalities in
(REC1334). All participants gave informed consent before taking part in the study. childhood and adolescent body-­mass index, weight, and height from
1953 to 2015: an analysis of four longitudinal, observational, British
Provenance and peer review Not commissioned; externally peer reviewed. birth cohort studies. Lancet Public Health 2018;3:e194–203.
Data availability statement Data are available upon reasonable request. NSHD 17 Elhakeem A, Cooper R, Bann D, et al. Childhood socioeconomic
data are available online (https://www.​nshd.​mrc.​ac.​uk/​data). Data from the position and adult leisure-­time physical activity: a systematic review.
remaining cohorts are available from the UK Data Archive (https://www.​data-​ Int J Behav Nutr Phys Act 2015;12:92.
18 Brant R. Assessing proportionality in the proportional odds model for
archive.​ac.​uk).
ordinal logistic regression. Biometrics 1990;46:1171–8.
Supplemental material This content has been supplied by the author(s). It has 19 Bann D, Villadsen A, Maddock J. Changes in the behavioural
not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been determinants of health during the coronavirus (COVID-19) pandemic:
peer-­reviewed. Any opinions or recommendations discussed are solely those gender socioeconomic and ethnic inequalities in 5 British cohort
studies. medRxiv 2020.
of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and
20 Silverwood RJ, Calderwood L, Sakshaug JW. A data driven approach
responsibility arising from any reliance placed on the content. Where the content to understanding and handling nonresponse in the next steps cohort.
includes any translated material, BMJ does not warrant the accuracy and reliability London: UCL Centre for Longitudinal Studies: CLS Working Paper,
of the translations (including but not limited to local regulations, clinical guidelines, 2020.
terminology, drug names and drug dosages), and is not responsible for any error 21 Mostafa T, Narayanan M, Pongiglione B, et al. Improving the
and/or omissions arising from translation and adaptation or otherwise. plausibility of the missing at random assumption in the 1958 British
birth cohort: a pragmatic data driven approach. cls working paper
Open access This is an open access article distributed in accordance with the 2020/6. London: UCL Centre for Longitudinal Studies, 2020.
Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits 22 Gordon EH, Peel NM, Samanta M, et al. Sex differences in frailty: a
others to copy, redistribute, remix, transform and build upon this work for any systematic review and meta-­analysis. Exp Gerontol 2017;89:30–40.
purpose, provided the original work is properly cited, a link to the licence is given, 23 Sterne JAC, Sutton AJ, Ioannidis JPA, et al. Recommendations for
and indication of whether changes were made. See: https://​creativecommons.​org/​ examining and interpreting funnel plot asymmetry in meta-­analyses
licenses/​by/​4.​0/. of randomised controlled trials. BMJ 2011;343:d4002–d02.
24 Orsini N, Li R, Wolk A, et al. Meta-­Analysis for linear and nonlinear
dose-­response relations: examples, an evaluation of approximations,
ORCID iD
and software. Am J Epidemiol 2012;175:66–73.
Constantin-­Cristian Topriceanu http://​orcid.​org/​0000-​0001-​5826-​9617 25 Pedersen AB, Mikkelsen EM, Cronin-­Fenton D, et al. Missing data
and multiple imputation in clinical epidemiological research. Clin
Epidemiol 2017;9:157–66.
26 Grumbach K. Chronic illness, comorbidities, and the need for
REFERENCES medical generalism. Ann Fam Med 2003;1:4–7.
1 Bleich SN, Jarlenski MP, Bell CN, et al. Health inequalities: trends, 27 Bayliss EA, Steiner JF, Fernald DH, et al. Descriptions of barriers to
progress, and policy. Annu Rev Public Health 2012;33:7–40. self-­care by persons with comorbid chronic diseases. Ann Fam Med
2 Kojima G, Liljas AEM, Iliffe S. Frailty syndrome: implications and 2003;1:15–21.
challenges for health care policy. Risk Manag Healthc Policy 28 Valderas JM, Starfield B, Sibbald B, et al. Defining comorbidity:
2019;12:23–30. implications for understanding health and health services. Ann Fam
3 Mein SA. COVID-19 and Health Disparities: the Reality of “the Great Med 2009;7:357–63.
Equalizer”. J Gen Intern Med 2020;35:2439–40. 29 Solé-Auró A, Crimmins EM. Who cares? A comparison of informal
4 Aspell N, O'Sullivan M, O'Shea E, apos SM, et al. Predicting and formal care provision in Spain, England and the USA. Ageing
admission to long-­term care and mortality among community-­ Soc 2014;34:495–517.
based, dependent older people in Ireland. Int J Geriatr Psychiatry 30 Cookson R, Propper C, Asaria M. Socio-­Economic inequalities in
2019;34:999–1007. health care in England. Fisc Stud 2016;37:371–403.
5 Kuh D, Pierce M, Adams J, et al. Cohort profile: updating the cohort 31 He D, Dushoff J, Day T, et al. Inferring the causes of the three waves
profile for the MRC national survey of health and development: a of the 1918 influenza pandemic in England and Wales. Proc Biol Sci
new clinic-­based data collection for ageing research. Int J Epidemiol 2013;280:20131345.
2011;40:e1–9. 32 Maringe C, Spicer J, Morris M, et al. The impact of the COVID-19
6 Power C, Elliott J. Cohort profile: 1958 British birth cohort (National pandemic on cancer deaths due to delays in diagnosis in England,
child development study). Int J Epidemiol 2006;35:34–41. UK: a national, population-­based, modelling study. Lancet Oncol
7 Elliott J, Shepherd P. Cohort profile: 1970 British birth cohort 2020;21:1023–34.
(BCS70). Int J Epidemiol 2006;35:836–43. 33 Khatana SAM, Groeneveld PW. Health disparities and the
8 Calderwood L, Sanchez C. Next steps (formerly known as the coronavirus disease 2019 (COVID-19) pandemic in the USA. J Gen
longitudinal study of young people in England). Open Health Data Intern Med 2020;35:2431–2.
2016;4:e2 34 Beaunoyer E, Dupéré S, Guitton MJ. COVID-19 and digital
9 Connelly R, Platt L. Cohort profile: UK millennium cohort study (mcs). inequalities: reciprocal impacts and mitigation strategies. Comput
Int J Epidemiol 2014;43:1719–25. Human Behav 2020;111:106424.
10 Viertiö S, Kiviruusu O, Piirtola M, et al. Gender and psychological 35 Crawford A, Serhal E. Digital Health Equity and COVID-19: The
distress: contribution of work-­family balance. Eur J Public Health Innovation Curve Cannot Reinforce the Social Gradient of Health. J
2019;29. Med Internet Res 2020;22:e19361.
11 Montazeri A, Harirchi AM, Shariati M, et al. The 12-­Item general 36 Shepherd BE, Rebeiro PF, Caribbean, Central and South America
health questionnaire (GHQ-12): translation and validation study of the network for HIV epidemiology. Brief report: assessing and
Iranian version. Health Qual Life Outcomes 2003;1:66 interpreting the association between continuous covariates and
12 Bowling A, Pikhartova J, Dodgeon B. Is mid-­life social participation outcomes in observational studies of HIV using splines. J Acquir
associated with cognitive function at age 50? results from the Immune Defic Syndr 2017;74:e60–3.

10 Topriceanu C-­C, et al. BMJ Open 2021;11:e045813. doi:10.1136/bmjopen-2020-045813

 © 2021 Author(s) (or their employer(s)) 2021. Re-use permitted under CC
BY. Published by BMJ. This is an open access article distributed in
accordance with the Creative Commons Attribution 4.0 Unported (CC BY
4.0) license, which permits others to copy, redistribute, remix, transform and
build upon this work for any purpose, provided the original work is properly
cited, a link to the licence is given, and indication of whether changes were
made. See: https://creativecommons.org/licenses/by/4.0/. Notwithstanding
the ProQuest Terms and Conditions, you may use this content in accordance
with the terms of the License.