Factors associated with enrolment into a national COVID-19 pulse oximetry remote monitoring programme in England: a retrospective observational study

Background Hypoxaemia is an important predictor of severity in individuals with COVID-19 and can present without symptoms. The COVID Oximetry @home (CO@h) programme was implemented across England in November, 2020, providing pulse oximeters to higher-risk people with COVID-19 to enable early detection of deterioration and the need for escalation of care. We aimed to describe the clinical and demographic characteristics of individuals enrolled onto the programme and to assess whether there were any inequalities in enrolment. Methods This retrospective observational study was based on data from a cohort of people resident in England recorded as having a positive COVID-19 test between Oct 1, 2020, and May 3, 2021. The proportion of participants enrolled onto the CO@h programmes in the 7 days before and 28 days after a positive COVID-19 test was calculated for each clinical commissioning group (CCG) in England. Two-level hierarchical multivariable logistic regression with random intercepts for each CCG was run to identify factors predictive of being enrolled onto the CO@h programme. Findings CO@h programme sites were reported by NHS England as becoming operational between Nov 21 and Dec 31, 2020. 1 227 405 people resident in 72 CCGs had a positive COVID-19 test between the date of programme implementation and May 3, 2021, of whom 19 932 (1·6%) were enrolled onto the CO@h programme. Of those enrolled, 14 441 (72·5%) were aged 50 years or older or were identified as clinically extremely vulnerable (ie, having a high-risk medical condition). Higher odds of enrolment onto the CO@h programme were found in older individuals (adjusted odds ratio 2·21 [95% CI 2·19–2·23], p<0·001, for those aged 50–64 years; 3·48 [3·33–3·63], p<0·001, for those aged 65–79 years; and 2·50 [2·34–2·68], p<0·001, for those aged ≥80 years), in individuals of non-White ethnicity (1·35 [1·28–1·43], p<0·001, for Asian individuals; 1·13 [1·04–1·22], p=0·005, for Black individuals; and 1·17 [1·03–1·32], p=0·015, for those of mixed ethnicity), in those who were overweight (1·31 [1·26–1·37], p<0·001) or obese (1·69 [1·63–1·77], p<0·001), or in those identified as clinically extremely vulnerable (1·58 [1·51–1·65], p<0·001), and lower odds were reported in those from the least socioeconomically deprived areas compared with those from the most socioeconomically deprived areas (0·75 [0·69–0·81]; p<0·001). Interpretation Nationally, uptake of the CO@h programme was low, with clinical judgment used to determine eligibility. Preferential enrolment onto the pulse oximetry monitoring programme was observed in people known to be at the highest risk of developing severe COVID-19. Funding NHS England, National Institute for Health Research, and The Wellcome Trust.


Introduction
The COVID-19 pandemic led to increased use of digital technologies in the delivery of clinical care. 1,2 In particular, remote monitoring devices played a role in the delivery of care to patients in their homes. [3][4][5] Hypoxaemia (a low arterial oxygen concentration) indicates a need for hospital admission, yet many patients with COVID-19 could have silent hypoxaemia with few symptoms of low blood oxygen concentrations. 6,7 Measurement of blood oxygen saturation was therefore recognised as a crucial part of clinical assessment. 4 The COVID Oximetry @home (CO@h) programme was launched by National Health Service (NHS) England in November, 2020, to provide pulse oximeters to highrisk people diagnosed with COVID-19 to support early recognition of hypoxaemia. 8,9 Further information about the CO@h programme is available online. 8,9 Adults either suspected or known to have COVID-19 and aged 65 years or older (later extended to those aged >50 years), or identified as clinically extremely vulnerable were eligible for enrolment onto the CO@h programme. 8 Additionally, clinical judgment could be applied to include other risk factors. 10 Those enrolled were asked to e195 www.thelancet.com/digital-health Vol 5 April 2023 record daily pulse oximetry readings and to contact primary care services for readings of 93-94% or emergency services for readings of 92% or less. 8 Sites differed in how readings were recorded and reported (eg, electronically or by use of paper diaries) and in the frequency of contact from health-care staff. 11 The service built upon remote monitoring services provided by individual clinical commissioning groups (CCGs; NHS organisations that commission local NHS services in England) and hospital trusts earlier in the pandemic. 8,12 Previous studies have identified several groups to be at increased risk of mortality due to COVID-19, including older individuals, those of non-White ethnicity, men, and those living in more socioeconomically deprived areas. Substantial differences have been found to persist after accounting for a broad range of clinical risk factors. 13,14 Additionally, several long-term conditions are strongly associated with an increased risk of complications due to COVID-19, including immunosuppression, cancer, and diabetes. 13,14 Although telemedicine and remote monitoring have offered the opportunity for many individuals to receive care at home during the pandemic, concerns have been raised that not all groups benefit equally from these innovations. 15 Many of the patient-related factors associated with barriers to the effective use of tele medicine or remote monitoring, including older age, socioeconomic deprivation, and multimorbidity, are also associated with increased risk of mortality from COVID-19. [13][14][15][16][17] It is therefore important to understand potential inequalities in the provision of remote monitoring programmes for COVID-19 and to identify whether factors predisposing individuals to enrolment align with previous understanding of the clinical risk posed by  We aimed to describe the clinical and demographic characteristics of individuals enrolled onto the national CO@h programme in England and to identify whether any inequalities in enrolment exist.

Data sources and processing
This retrospective observational study used data from a cohort of people resident in England recorded as having a positive COVID-19 test between Oct 1, 2020, and May 3, 2021. This study used datasets linked and provided by NHS Digital as part of an evaluation of NHS England's CO@h programme. Data on positive COVID-19 tests were provided through the Public Health England Second Generation Surveillance System, which includes COVID-19 test results conducted in laboratories across England. 18 Positive PCR tests and lateral flow device tests were included. Where more than one positive test was identified for each participant, the date of the first test was used.

Research in context
Evidence before this study This study was conducted alongside a separate systematic review published in The Lancet Digital Health in 2022, which appraised the available evidence to date on the safety and effectiveness of COVID-19 pulse oximetry remote monitoring programmes across the world. This systematic review searched five databases (MEDLINE, Embase, Global Health, medRxiv, and bioRxiv) from database inception to April 15, 2021, and included feasibility studies, clinical trials, and observational studies, as well as preprints. 561 studies were found, of which 13 were included in a narrative synthesis. Studies were included if they targeted adult patients with confirmed or presumptive COVID-19; used a pulse oximeter as a remote monitoring tool for SpO₂; involved any new or existing remote patient monitoring system at home, in community settings, or both; and assessed the effectiveness and safety of pulse oximetry in these settings. Studies were excluded if they did not focus on COVID-19; if they focused on patients admitted to hospital; if they did not use hand-held pulse oximetry; if they used pulse oximetry with only a proportion of study participants, and results for this subgroup were not reported individually; and if they assessed the accuracy of a specific pulse oximeter type or brand, without monitoring patient deterioration or health-care service use. The systematic review found pulse oximetry remote patient monitoring to be a safe way to monitor patients with COVID-19 at home and to identify the risk of deterioration and the need for advanced care. However, the systematic review did not investigate the factors predictive of enrolment onto such programmes.

Added value of this study
This study is a retrospective, national, multisite observational study of more than 1·2 million individuals with a positive COVID-19 test, examining the factors associated with enrolment onto the COVID Oximetry @home (CO@h) programme in England. As such, this study provides an understanding of how new remote monitoring pathways are implemented at scale and which groups are more or less likely to receive the intervention. We found that those with the highest clinical risk of COVID-19 were generally more likely to be enrolled into the CO@h programme. Lower odds of enrolment in some groups, including individuals with dementia or care home residents, might indicate potential barriers to the use of remote monitoring pathways by these groups.

Implications of all the available evidence
Remote patient monitoring with pulse oximetry has an important role in the management of individuals with COVID-19. Clinician autonomy was widely used to determine eligibility for the CO@h programme. Similar pathways should continue to provide this important flexibility for clinicians to match patients to pathways outside of stricter eligibility criteria.
Data on the date and site of enrolment of participants into the CO@h programme were submitted from participating sites to NHS Digital's Strategic Data Collection Service. 19 The study population was assigned to a CCG based on a person's CCG of residence when the test was done. The CCG configuration as of April, 2021, was used. 20 The date of implementation of the CO@h programme in each CCG was provided by NHS England.
Primary care data were received from the General Practice Extraction Service Data for Pandemic Planning and Research (GDPPR). 21 Participant demo graphic data were derived from GDPPR, or where missing, from Hospital Episode Statistics data or the Emergency Care Data Set. 22,23 The lower layer super-output area (LSOA) of residence of each participant was linked to indices of relative socioeconomic deprivation by use of deciles of the Index of Multiple Deprivation (IMD) 2019. 24 Residence in a care home, BMI, and smoking status were derived from GDPPR. BMI was categorised as under weight (<18·5 kg/m²), healthy bodyweight (18·5-24·9 kg/m²), overweight (25·0-29·9 kg/m²), and obese (≥30·0 kg/m²). Clinically extremely vulnerable status was recorded via a flag in the GDPPR record if a participant was identified on the NHS Digital Shielded Patient List (see appendix 1 p 1 for risk criteria).
Information on the following 12 chronic conditions was extracted from GDPPR based on Systematised Nomenclature of Medicine Clinical Terms (SNOMED-CT) codes: hypertension, chronic cardiac disease, chronic kidney disease, chronic respiratory disease, dementia, diabetes, chronic neurological disease (including epilepsy), learning disability, malignancy or immunosuppression, severe mental illness, peripheral vascular disease, and stroke or transient ischaemic attack. Only diagnoses recorded before the date of a positive COVID-19 test were included. Further details on data curation, and a list of included SNOMED-CT codes, are provided in appendix 1 (pp 1-7) and in appendix 2. Datasets were linked using a deidentified NHS patient ID.

Statistical analysis
The proportion of people enrolled onto the CO@h programme between the 7 days before and 28 days after a positive COVID-19 test was calculated for each CCG. This range was chosen to include individuals enrolled on the basis of clinical suspicion of COVID-19 before a positive result was reported, and to include those enrolled on the programme later in the period after their positive test. To exclude CCGs that had not implemented the CO@h programme or submitted enrolment data through the Strategic Data Collection Service, the analysis was restricted to CCGs in which at least 0·1% of all residents with a positive COVID-19 test were enrolled onto the CO@h programme.
Two-level hierarchical multivariable logistic regression was done to identify factors predictive of being enrolled onto the CO@h programme, with CCG of residence included as a random intercept. Separate two-level hierarchical multivariable logistic regression models were run for each of four age groups (18-49 years, 50-64 years, 65-79 years, and ≥80 years) and for CCGs in which up to 0·1% to less than 1·0%, 1·0% to less than 2·0%, 2·0% to less than 5·0%, and 5·0% or more residents with a positive COVID-19 test were enrolled. We also conducted two separate sensitivity analyses excluding clinically extremely vulnerable status and comorbidities.
This work was conducted as a national service evaluation of the CO@h programme, approved by Imperial College Health Trust on Dec 3, 2020. Data access was approved by the Independent Group Advising on the Release of Data (DARS-NIC-421524-R0Y3P) on April 15, 2021. The need for individual participant consent was waived.
Participants or the public were not involved in the design, conduct, or reporting of this research.

Role of the funding source
The study funders did not play a role in study design; in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the Article for publication. Researchers were independent from funders.

Results
CO@h programme sites were reported by NHS England as becoming operational between Nov 21 and Dec 31, 2020. 1 684 260 people resident in 106 CCGs in England had a positive COVID-19 test between the date of implementation of the CO@h programme in their CCG and May 3, 2021. Of these, 19 980 (1·2%) were enrolled onto the CO@h programme between 7 days before and 28 days after their positive COVID-19 test. In 21 CCGs, no data were received on residents recorded as being enrolled onto the CO@h programme. In a further 13 CCGs, fewer than 0·1% of all people with a positive COVID-19 test were recoded as being enrolled onto the CO@h programme. Collectively, these 34 CCGs accounted for 456 855 positive COVID- 19  their urbanicity, proportion of residents aged 65 years or older, and proportion of residents of White ethnicity. Included CCGs were significantly more likely to contain socioeconomically deprived areas than CCGs that were excluded (appendix 1 p 10). Table 1 shows the demographic and clinical characteristics of those with a positive COVID-19 test included in the analysis. Of those enrolled, 11 509 (57·7%) were aged 50-79 years. More women were enrolled than men (55·1% vs 44·8%). The majority (76·0%) of those enrolled were of White ethnicity. 11 829 (59·3%) of those enrolled were resident in the most socioeconomically deprived half of LSOAs in England (ie, IMD deciles 1-5). Of those enrolled, 4415 (22·2%) were identified as clinically extremely vulnerable, 811 (4·1%) were resident in a care home, 13 915 (69·8%) were either overweight or obese, and 10 691 (53·6%) were never-smokers. Chronic respiratory disease (32·9%), hypertension (30·8%), and diabetes (18·9%; type 1 and type 2) were the most common comorbidities. With reference to the CO@h programme eligibility criteria, 14 441 (72·5%) of those enrolled were aged 50 years or older or were identified as clinically extremely vulnerable. There was some correlation between ethnicity, age, and socioeconomic deprivation in the study population: 62 154 (91·2%) of those aged 80 years or older were of White ethnicity, compared to 492 628 (70·1%) of those younger than 50 years. Similarly, 69·6% of residents of the most

Clinically extremely vulnerable
Chronic neurological disease (including epilepsy)   Table 3 shows the estimated adjusted odds ratios for enrolment onto the CO@h programme for patient-level predictors from multivariable mixed-effects logistic regression models stratified by age group. Being identified as clinically extremely vulnerable, or overweight or obese, was associated with a significantly higher odds of enrolment in all age groups, with the highest odds ratios among those aged younger 50 years. In older age groups, clinical comorbidities were not significantly associated with a higher odds of enrolment except for chronic respiratory disease; however, among those aged 80 years or older, men were significantly more likely to be enrolled than women. Table 4 shows the proportion of patients enrolled onto the CO@h programme for different CCG levels of overall enrolment. The geographical location of each CCG according to the level of programme enrolment is shown in the figure, showing variation in uptake of the programme across the country, with higher enrolment in the northwest of England and part of the southwest of England. Table 5 shows the estimated adjusted odds ratios for enrolment onto the CO@h programme for patientlevel predictors from multivariable mixed-effects logistic regression models according to CCG levels of enrol ment. Across all categories of enrolment, higher age was a significant predictor of enrolment compared to those younger than 50 years. People of Asian ethnicity were significantly more likely to be enrolled than those of White ethnicity in CCGs enrolling more than 1% of people. In no cases were people of non-White ethnicity less likely to be enrolled than White people.
Across all categories of enrolment, people resident in the most deprived decile of LSOAs were more likely to be enrolled than those resident in less deprived LSOAs. Across all levels of enrolment, those identified as clinically extremely vulnerable, overweight, or obese

test enrolled onto the CO@h programme in each CCG in England
Areas shaded in grey correspond to excluded CCGs where less than 0·1% of individuals were enrolled. CCG=clinical commissioning group. CO@h=COVID Oximetry @home. ≥0·1%   were significantly more likely to be enrolled than those with a healthy bodyweight or those who were not identified as clinically extremely vulnerable.

Discussion
This study examined the characteristics of those enrolled into the national CO@h programme in England and identified features predictive of enrolment. Only 1·6% of all patients testing positive for COVID-19 were recorded as being enrolled onto the CO@h programme, which is low relative to what the study team and service providers were expecting. It is not possible to determine from the available data whether this apparent low rate of enrolment is a result of incomplete or missing data submitted from CO@h sites or whether it reflects genuine low rates of uptake. In either case, it is likely that those enrolled represent a minority of those eligible for the CO@h programme overall. Although 72·5% of those enrolled into the CO@h programme were 50 years or older or clinically extremely Mixed or multiple ethnic groups 10 (least deprived) 0·56 (0·46-0·68) <0·001 0·84 (0·64-1·10) 0·203 0·78 (0·70-0·87) <0·001 0·86 (0·73-1·01) 0·074

Smoking status
Never smoker (reference) ·· ·· ·· ·· Current smoker 0·87 (0·78-0·97) 0·009 0·79 (0·69-0·90) <0·001 0·96 (0·90-1·02) 0·229 0·97 (0·88-1·07) 0·553 (Table 5 continues on next page) vulnerable, more than one quarter were not, suggesting an important role for clinical discretion in enrolment. Many of the clinical and demographic characteristics associated with increased odds of enrolment into the CO@h programme have also been identified in previous studies as being associated with an increased risk of hospital admission or mortality due to COVID-19. 13,14 Patients of non-White ethnicity and those resident in the most socioeconomically deprived areas were significantly more likely to be enrolled, suggesting that these factors were incorporated into clinical judgment despite neither being explicitly part of national eligibility criteria.
People with dementia, a previous stroke or transient ischaemic attack, or chronic kidney disease were significantly less likely to be enrolled, after controlling for other clinical and demographic features. Patients with dementia could experience difficulties in the use of digital health technologies and might require the help of more digitally literate caregivers to effectively use these programmes. 25 There is a risk that inequalities in access to new clinical pathways, such as the CO@h programme, could emerge as a result of the availability and digital literacy of a patient's support network.
Although the present study found that care home residents were less likely to be enrolled into the CO@h programme, remote monitoring with pulse oximetry was used in care homes outside of CO@h, which were not included in this study. 26 The characteristics associated with enrolment varied across age groups. In individuals younger than 65 years, having one of several clinical comorbidities, high BMI, or being identified as clinically extremely vulnerable were particularly strongly predictive of enrolment. In individuals aged 65 years and older, these features were either less strongly associated with enrolment, or were not significantly associated at all. This finding suggests in part that the role of age alone is a sufficient criterion for enrolment. Chronic respiratory disease was the only clinical comorbidity significantly associated with enrolment across all age groups, emphasising its perceived importance as a risk factor for severe COVID-19. The finding that men were significantly more likely than women to be enrolled among those aged 80 years or older might reflect the view that men were at higher risk of severe COVID-19 than women. 13,14 The similarity of findings across different levels of uptake within CCGs suggests that similar criteria for enrolment into the programme operated independently of the scale at which the programme was implemented.
A comparison of the demographic and clinical characteristics shown in table 1 with the results of the multivariable models indicates some confounding. For example, 76·0% of those enrolled were of White ethnicity, compared to only 70·0% of all people with a positive COVID-19 test, whereas the odds of enrolment for those of White ethnicity were significantly lower than those of non-White ethnicity. This discrepancy might be partly explained by the correlation between ethnicity, age, and socioeconomic deprivation in the study population: 91·2% of those aged 80 years or older were of White ethnicity, compared to only 70·1% of those younger than 50 years. Similarly, 69·6% of residents of the most socioeconomically deprived decile of LSOAs in England were of White ethnicity compared to 88·7% in the least deprived decile of LSOAs.
This study brings together several primary, secondary, and public health datasets to provide a more detailed

Comorbidities
Hypertension Chronic neurological disease (including epilepsy) CO@h=COVID Oximetry @home. OR=odds ratio. CCG=clinical commissioning group. IMD=index of multiple deprivation. understanding of the clinical and demographic features associated with enrolment into the CO@h programme. Low recorded rates of enrolment might suggest either missing data from sites or substantial barriers to implementation at scale. Neither of these issues is likely to be random and this could have biased our results. It is likely that variation in enrolment between CCGs might partly result from difficulties associated with the implementation of the programme itself, including increased workload on health-care providers, lack of awareness of the programme, or uncertainty of the value of the programme compared to other methods of care.
It was also not possible to ascertain the clinical status of a patient (such as symptoms or oxygen saturation) at the time of enrolment and therefore not possible to account for the role of clinical acuity in enrolment decisions.
Enrolment processes varied across different sites and there was no unified method of enrolment. Additionally, methods of enrolment and criteria for enrolment could have changed over time during the study period. In all cases, clinical judgment was encouraged. Enrolment onto the CO@h programme might not have been solely reliant on the decision of a clinician. For example, individuals who perceive their own risk of severe COVID-19 to be high or who have greater trust in the NHS might be more likely to approach providers of the service or accept enrolment when offered. This study cannot quantify the extent to which clinical decision making, or the preferences and actions of individuals, could have contributed to the associations observed. It is also possible that individuals with their own pulse oximeters might have monitored their own condition without enrolment onto a CO@h programme, or were monitored by clinicians outside of the programme. If more affluent patients were more likely to own a pulse oximeter, this might explain some of the observed trends in enrolment by area-level socioeconomic deprivation.
Moreover, enrolment provides no indication of whether an individual was able to successfully perform the monitoring and reporting required, or whether the CO@h programme pathway was clinically beneficial. Two further analyses conducted as part of the wider evaluation of the CO@h programme found no considerable impact of the programme on all-cause mortality or hospital admission in individuals who had tested positive for COVID-19 at a population level, but did find lower odds of mortality in a subgroup of patients clinically assessed in emergency departments. 27,28 This study found that those most at risk from severe COVID-19 were more likely to be enrolled into the CO@h programme. As such, enrolment into the programme reflected preferential allocation according to contemporaneous knowledge of the factors associated with risk of severe COVID-19.
Although it is not possible to ascertain whether differential enrolment relates to perceived risk by a clinician or differences in patients' health-seeking behaviour, our findings suggest that clinical judgment might have played an important role in enabling more nuanced incorporation of known risk factors for severe COVID-19 in enrolment decisions than age or criteria based on clinically extremely vulnerable status would permit. Future similar programmes should continue to encourage and facilitate clinical judgment in enrolment decisions.
The finding that some of the frailest individuals and care home residents were less likely to be enrolled is concerning, and further research is needed to understand whether this observation results from the provision of care by alternative pathways, or whether it reflects reduced access to remote monitoring pathways in these groups.
Although provision appeared to be equitable within individual sites, wide variation was observed in the adoption of the programme across the country. More research is needed to understand why some sites enrolled a larger proportion of participants than others to help guide future national level monitoring programmes.
In conclusion, the CO@h programme in England was implemented to varying extents across CCGs in England; however, only a minority of eligible people appear to have been enrolled. Of those recorded as enrolled, older people, those of non-White ethnicity, those from more socioeconomically deprived areas, or diagnosed with one of several chronic medical conditions were significantly more likely to be enrolled. This finding indicates that the service was preferentially used by those known to be at increased clinical risk.