Chronic pain and COVID-19 hospitalisation and mortality: a UK Biobank cohort study

Supplemental Digital Content is Available in the Text. Chronic pain is associated with increased risk of hospitalisation due to COVID-19, and this risk increases with the number of pain sites.


Introduction
Chronic pain significantly affects well-being, with chronic low back pain being the leading cause of disability globally. 15 In the United Kingdom, the prevalence of chronic pain in adults is estimated to be between 35.0% and 51.3%. 11 Globally, chronic pain affects 1 in 5 adults, although this is believed to be an underestimate. 22 Chronic pain is a complex condition which is challenging to treat. 32,35 The burden of disease due to chronic pain and demand for effective pain management is likely to increase during the COVID-19 pandemic, with joint and muscle pain being reported both in the acute phase and by those with long COVID. 3,30 Furthermore, pre-existing musculoskeletal pain can be exacerbated by COVID-19. 12 Several authors have commented on the implications of COVID-19 for pain management 7,42,44 including the use of telemedicine to minimise face-to-face encounters and reduce SARS-CoV-2 transmission. 9,16,29,47 Although it is clear that chronic pain can be a result of SARS-CoV-2 infection, it remains unclear whether chronic pain may predispose to more severe SARS-CoV-2 infection or higher risk of adverse outcome.
To mitigate the impact of COVID-19 and target public health and clinical interventions, many hospital and community studies have tried to identify risk factors for COVID-19 outcomes. Sociodemographic characteristics such as older age, socioeconomic deprivation, male sex, and Black and South Asian ethnicity have consistently been associated with higher risk of SARS-CoV-2 infection and death in the United Kingdom. 21,34,40,51 The association between ethnicity and COVID-19 has been shown to be modified by occupation, although other factors including overcrowded housing, poorer socioeconomic status, and structural racism are also important. 18,26,33,37 Some individual long-term conditions (LTCs), for example, cardiac disease, diabetes, and obesity, have been associated with higher COVID-19 case fatality. 8,34 Furthermore, multimorbidity (the presence of 2 or more LTCs) is also associated with increased risk of SARS-CoV-2 infection 34 as is frailty. 41 Chronic pain is more likely to be reported by those of Asian and Black ethnicity 39 and in those of low socioeconomic status. 20 A study of multimorbidity in primary care found that pain was associated with all the 10 most prevalent multimorbidity clusters. 4 Research into whether chronic pain is associated with SARS-CoV-2 infection or COVID-19 mortality is limited. A recent study observed that, in children and adults with sickle cell disease, those with frequent prior acute care visits for pain were more likely to be hospitalised during their COVID-19 illness than those without such a history. 36 However, it is not known whether this relationship exists in the general population.
We explored whether UK Biobank participants with selfreported chronic pain had a higher risk of COVID-19 hospitalisation and mortality.

Participants
UK Biobank is a prospective cohort study. It recruited 502,624 participants aged 37 to 73 years across England, Scotland, and Wales between 2006 and 2010. At baseline, touch-screen questionnaires and nurse-administered interviews were used to collect demographic, health, environmental, and lifestyle data using standardised protocols and to record biological measurements. 6,45 This study received ethical approval from the North West Multi-Centre Research Ethics Committee (REC reference: 16/NW/0274). All participants gave written informed consent for data collection, analysis, and record linkage.

Definitions
Baseline data from UK Biobank were linked, both prospectively and retrospectively, to hospital inpatient data, comprising Hospital Episode Statistics (HES) Admitted Patient Care data for England and Scottish Morbidity Record (SMR01) data for Scotland. Potential participants were excluded if their baseline assessment centre was in Wales because Welsh hospital inpatient data have not been updated to include the pandemic period.
Primary outcome was COVID-19 hospitalisation, defined as ICD code U071 (virus identified). Secondary outcome was COVID-19 mortality, defined as ICD code U071 or U072 (virus probable, suspected, or clinically epidemiologically diagnosed) recorded as the primary cause of death.
The hospital data included admission date and COVID-19 diagnosis. For those who had a COVID-19 diagnosis, admissions data were available between 15 of August 2019 and 26 of March 2021. Data were linked to Death Register Data for England and Scotland (available from 5 March 2020 to 17 March 2021). All COVID-19 deaths were included regardless of where they occurred: in the community and during or after admission to hospital.
Chronic pain, self-reported at baseline (2006-2010), using a touch-screen questionnaire, was defined as having pain for at least 3 months in one or more of the following sites: head, face, neck/ shoulder, back, abdomen, hip, and knee. A binary yes/no variable was constructed for chronic pain. The total number of sites of chronic pain was categorised as none (no chronic pain), 1, 2 to 3, or 4 to 7. Chronic widespread pain was defined as self-reported pain all over the body for at least 3 months. It was analysed as a separate exposure, comparing participants who reported chronic widespread pain with those without chronic pain.
Ethnicity was self-reported and categorised as White, Black, South Asian, Chinese, mixed, or other. Current age was calculated as age at assessment plus the number of years between assessment date and date of death or censor date. Current age was entered into all analyses as a continuous variable. Body mass index (BMI) was derived from weight (kg)/ height (m) 2 and categorised into underweight (,18.5 kg/m 2 ), normal weight (18.5-24.9 kg/m 2 ), overweight ($25-29.9 kg/m 2 ), obese ($30-34.9 kg/m 2 ), and morbidly obese ($35 kg/m 2 ). 50 Area-level socioeconomic deprivation was assessed using the Townsend deprivation index, which incorporates unemployment, car ownership, home ownership, and household overcrowding. 48 Higher Townsend scores represent greater socioeconomic deprivation; scores were categorised into quintiles within the study sample. Smoking status, frequency of alcohol consumption, and physical activity were self-reported. Smoking status was categorised as never or current or previous smoker. Alcohol consumption was categorised into never or special occasions only, 1 to 3 times a month, 1 to 4 times a week, and daily or almost daily. Physical activity was categorised into none, low, medium, or high using Metabolic Equivalent Task (MET) scores based on the International Physical Activity Questionnaire (IPAQ short form) scoring protocol. 23 Number of LTCs was defined as a count of 43 self-reported conditions which have previously been used in UK Biobank studies of multimorbidity (Supplementary Table 1, available at http://links.lww.com/PAIN/B629). 24 This list of conditions is based on previously published literature on multimorbidity. 1,38 For chronic pain as a binary and ordinal variable, the "painful conditions" grouping contained only trigeminal neuralgia, shingles, and headache. For chronic widespread pain, the "painful conditions" grouping also contained back pain/problems, sciatica/disc/nerve problems, plantar fasciitis, carpal tunnel syndrome, and joint osteo/spine arthritis/spondylitis/arthritis with no other symptoms. LTC count was entered into models categorised as 0, 1, 2 to 3, and $4 LTCs.

Statistical analyses
Univariable Poisson regression analysis was performed for the association between chronic pain and COVID-19 diagnosis (in hospital). In multivariable analysis, the model was first adjusted for sociodemographic factors (current age, sex, Townsend deprivation quintile, ethnicity, and assessment centre location), then additionally adjusted for lifestyle factors (smoking status, alcohol frequency, BMI, and physical activity), and finally for number of LTCs. The adjustment for number of LTCs was performed last because LTCs can be considered as a confounder or a mediator 10,49 (or both) in the relationship between chronic pain and COVID-19 diagnosis and mortality. Chronic pain was coded first as a binary variable, then as an ordinal variable derived from number of pain sites (none, 1, 2-3, and 4-7), and finally as chronic widespread pain vs no chronic pain.
Cox proportional hazards regression was performed for the association between chronic pain and COVID-19 mortality (deaths in the whole study sample), and for the association between chronic pain and COVID-19 case fatality (deaths among those hospitalised for COVID-19), both adjusted as above. Models included time to event (mortality or end of follow-up) from baseline assessment. The proportional hazards assumption was assessed through formal tests of Schoenfeld residuals. Analyses were undertaken using Stata v14. admission for COVID-19, and a further 19,438 were excluded because they attended a Welsh assessment centre (Supplementary Fig. 1, available at http://links.lww.com/PAIN/B629). Of the remainder, complete data on chronic pain and covariates were available for 441,403 participants. Of these, 3180 (0.7%) were admitted with a COVID-19 diagnosis and there were 1040 (0.2%) COVID-19-related deaths. 1724 (54%) of those hospitalised for COVID-19 had a history of chronic pain, as did 539 (52%) of those whose death was related to COVID-19. Table 1 presents participant characteristics by COVID-19 hospitalisation. Among participants with a positive COVID-19 test in hospital, 54% had self-reported chronic pain in at least 1 site at baseline, compared with 43% of those without a positive test. Presence of chronic pain was associated with SARS-CoV-2 infection univariably ( Table 2; IRR 1.57, 95% CI 1.46-1.68, P , 0.001); the association was attenuated after adjustment for potential confounders (adjusted IRR 1.25, 95% CI 1.17-1.35; P , 0.001) and further attenuated but remained significant after adjustment for LTCs (fully adjusted IRR 1.16, 95% CI 1.08-1.24; P , 0.001).
There was no clear association between chronic pain and COVID-19 case fatality. For the exposures, chronic pain status ( Table 2) and number of pain sites ( Table 3) hazard ratios were close to 1. For chronic widespread pain, there was a suggestion of increased case fatality but 95% confidence intervals were very wide ( Table 4; fully adjusted HR 1.34, 95% CI 0.90-1.99, P-value50.150). This may reflect the rarity of the outcome rather than there being no effect.

Discussion
To the best of our knowledge, this is the first study to explore the associations between chronic pain and COVID-19 hospitalisation and mortality in the general population. Chronic pain, both at The effect sizes observed imply potentially small increases in the risk of COVID-19 hospitalisation at a population level. For example, the lower limit of the fully adjusted 95% confidence interval when chronic pain is treated as a binary variable represents only a relative increased risk of 8% when compared with those without chronic pain. However, those with pain at 4 to 7 sites are between 28% and 74% more likely to have COVID-19 infection compared with those with no pain, with a best estimate of 49%. This is important in clinical terms.
The relationship between chronic pain and hospitalisation for COVID-19 infection is currently unclear. One possibility is that chronic pain is a proxy measure for unmeasured risk factors or undiagnosed disease. It has been linked to a wide range of pathophysiological consequences, including effects on mood, cognition, sleep, cardiovascular risk, quality of life, and general functioning. 13 Chronic widespread pain is associated with poorer self-reported general health, 46 and people with chronic pain have higher rates of anxiety and depression. 19 Alternatively, inflammatory mechanisms underpinning some chronic pain conditions 5 may also make individuals susceptible to more severe COVID-19 infection, requiring hospitalisation, as may also be the case for other conditions where inflammation seems to be causally related, eg, cardiovascular disease. 27 Furthermore, diminished physical functioning due to chronic pain may predispose to complications of COVID-19. For example, greater immobility due to chronic pain may potentiate thromboembolic complications and poorer preinfection cardiorespiratory function.
The strengths of UK Biobank include its large sample size and extensive phenotyping, which enabled us to adjust for a wide range of potential confounders including comorbidities and sociodemographic and lifestyle risk factors. Nonetheless, as with all observational studies, residual confounding because of unknown or unmeasured confounders is possible, for example, differential access to testing or hospitalisation. UK Biobank is not representative of the general population (response rate 5.5%) 14 Table 2 Poisson and Cox regression models of the associations between chronic pain status and COVID-19 hospitalisation, mortality, and case fatality.  Poisson and Cox regression models of the association between chronic pain category and COVID-19 hospitalisation, mortality, and case fatality. Data on chronic pain were collected before the COVID-19 pandemic excluding the possibility of reverse causation. In this analysis, we assumed that most of the UK Biobank participants who reported chronic pain at baseline continued to experience persistent pain over the medium to long term albeit that it may fluctuate regarding intensity and interference. 17 Although pain status may, in reality, have changed over the intervening period, this is unlikely to have introduced systematic error. Studies that measure duration of chronic pain show that it persists, at least in the medium term, for 5 to 7 years. 28,31,43 Kamaleri and colleagues showed that the number of pain sites (not specifically chronic pain) a person reported remained stable over a 14-year period 25 ; a similar time frame from UK Biobank baseline assessments to the COVID-19 pandemic (10-14 years). Therefore, any change in pain status over time is likely to take the form of missing data on new onset pain which would likely result in an underestimate of the real effect.
The availability of testing and the testing strategies used have changed over the course of the pandemic. Therefore, incomplete ascertainment of infections is possible. Our ascertainment method should be relatively robust against changes in test strategy because we limited outcomes to COVID-19 hospitalisations and mortality. This excludes people tested in the community who were likely to have had milder symptoms, were more likely to have been denied access to testing, and experienced more dramatic temporal changes in access to testing.

Conclusion
Our analyses of UK Biobank data demonstrated an association between chronic pain (at separate sites or all over the body) and COVID-19 hospitalisation, independent of known LTCs, and clear evidence of a dose-response relationship with number of pain sites. Pain may be a proxy measure of undiagnosed underlying physical or mental health disease processes. Future studies are needed to corroborate our novel findings, investigate the underlying mechanisms, and investigate whether pre-existing pain is also an independent risk factor for long COVID. Poisson and Cox regression models of the association between chronic widespread pain status and COVID-19 hospitalisation, mortality, and case fatality.