Examining patterns of multimorbidity, polypharmacy and risk of adverse drug reactions in chronic obstructive pulmonary disease: a cross-sectional UK Biobank study

Objective This study aims: (1) to describe the pattern and extent of multimorbidity and polypharmacy in UK Biobank participants with chronic obstructive pulmonary disease (COPD) and (2) to identify which comorbidities are associated with increased risk of adverse drug reactions (ADRs) resulting from polypharmacy. Design Cross-sectional. Setting Community cohort. Participants UK Biobank participants comparing self-reported COPD (n=8317) with no COPD (n=494 323). Outcomes Multimorbidity (≥4 conditions) and polypharmacy (≥5 medications) in participants with COPD versus those without. Risk of ADRs (taking ≥3 medications associated with falls, constipation, urinary retention, central nervous system (CNS) depression, bleeding or renal injury) in relation to the presence of COPD and individual comorbidities. Results Multimorbidity was more common in participants with COPD than those without (17% vs 4%). Polypharmacy was highly prevalent (52% with COPD taking ≥5 medications vs 18% in those without COPD). Adjusting for age, sex and socioeconomic status, those with COPD were significantly more likely than those without to be prescribed ≥3 medications contributing to falls (OR 2.27, 95% CI 2.13 to 2.42), constipation (OR 3.42, 95% CI 3.10 to 3.77), urinary retention (OR 3.38, 95% CI 2.94 to 3.87), CNS depression (OR 3.75, 95% CI 3.31 to 4.25), bleeding (OR 4.61, 95% CI 3.35 to 6.19) and renal injury (OR 2.22, 95% CI 1.86 to 2.62). Concomitant cardiovascular disease was associated with the greatest risk of taking ≥3 medications associated with falls/renal injury. Concomitant mental health conditions were most strongly associated with medications linked with CNS depression/urinary retention/bleeding. Conclusions Multimorbidity is common in COPD and associated with high levels of polypharmacy. Co-prescription of drugs with various ADRs is common. Future research should examine the effects on healthcare outcomes of co-prescribing multiple drugs with similar potential ADRs. Clinical guidelines should emphasise assessment of multimorbidity and ADR risk.


Multimorbidity and polypharmacy 240
Prevalence of each category of comorbidity was higher in those with COPD than without (table 2). 241 After controlling for age, sex and socioeconomic status, those with self-reported COPD were 242 significantly more likely than those without to have each category of comorbidity examined: 243 cardiovascular disease (OR 1.45; 95% confidence interval (CI) 1. 39 (Table 3). Findings were similar for GOLD COPD however, after adjusting for 280 additional potentially confounding variables, results for bleeding risk were not statistically significant 281 in this sensitivity analysis (Table 3). 282 Table 3. Odds ratios (with 95% CI) for taking 3 of more medications associated with similar ADRs ADR Self report COPD compared with no COPD N=502,640 Finally, each category of ADR risk was assessed in a subgroup analysis for each category of 284 comorbidity (cardiovascular, GI, cancer, mental health and painful conditions) comparing those with 285 and without COPD (e.g. participants with cardiovascular disease plus COPD compared with 286 participants with cardiovascular disease alone, etc.). These models were adjusted for age, sex and 287 socioeconomic status only. Within each category of comorbidity, those with self-reported COPD 288 were more likely to be at risk of each ADR than those without COPD (appendix 3). Not all results 289 were statistically significant when using GOLD COPD (Appendix 3).

Summary of main findings 297
Multimorbidity and polypharmacy in COPD were common among UK Biobank participants. The 298 presence of one or more comorbidity was highly prevalent in those with COPD (85%). More than half 299 reported polypharmacy (five or more medications), and 15% reported 10 or more medications. The 300 prevalence of cardiovascular disease, as well as the degree of polypharmacy, was higher among 301 those with more severe airflow obstruction. 302 For the first time, our data demonstrates that those with COPD were more likely than those without 303 to be prescribed multiple medications (≥ three) with similar ADRs. Those with COPD plus 304 cardiovascular comorbidity were most likely to be taking multiple medications with a risk of falls and 305 of renal injury, while those with COPD plus comorbid mental health conditions were most likely to 306 be taking medications causing constipation, CNS depression and bleeding. Within each category of 307 comorbidity, those with COPD were more likely to be taking multiple medications with similar ADRs 308 than those without. These associations between patterns of multimorbidity and specific ADR risks 309 have not been described or quantified previously. 310 311

Strengths and limitations 312
Strengths of this study include the large sample size with representation from different areas of the 313 UK. The range of data collected at UK Biobank assessment centres meant it was possible to compare 314 a range of sociodemographic characteristics as well as spirometry data, the latter being unusual for a 315 large community based cohort. It is recognised, however, that UK Biobank participants show some 316 evidence of 'healthy volunteer bias', differing from the UK average on a number of socioeconomic, 317 lifestyle and health-related measures. Specifically they are less socioeconomically deprived, less 318 likely to smoke, to be obese, and have fewer self-reported health conditions. ( 18 as well as medication data were self-reported, with no alternative means of verification. We 320 attempted to minimise this limitation by identifying a subset of those with COPD meeting the GOLD 321 diagnostic criteria and repeating the analyses with this subset. Importantly, spirometry values were 322 also pre-bronchodilator, which is in contravention to guidelines for diagnosing COPD. Additionally, 323 information was not available about the strength of indication for medications and individual 324 susceptibility to risk, which is a limitation when considering the risk of ADRs. 325 The use of the Scottish Government Polypharmacy Guideline allowed analysis of potential ADR risk 326 by specific common ADRs. The intended purpose of this guideline, however, was not to identify 327 potential risk from a population sample, but rather to identify potential causes of symptoms or 328 complications. The analysis in this study, therefore, serves only as an approximation of potential risk, 329 not an absolute marker of inappropriate polypharmacy. The cross-sectional nature of this analysis 330 also precludes an analysis of actual harm as a result of polypharmacy. Despite these limitations, 331 however, the co-prescription of multiple medications with similar ADRs strongly implies greater 332 potential for harm. The association of such prescribing patterns with COPD, across a range of 333 potential ADRs, is clear from our findings. This analysis is, to the author's knowledge, the first to 334 attempt to quantify this risk for specific ADRs in this way. 335 336

Context and implications 337
The increased prevalence of individual comorbid conditions such as coronary heart disease, 338 hypertension, diabetes, dyspepsia, osteoporosis, cancer, depression and anxiety in those with COPD 339 is similar to the findings from other population based studies of comorbidities in COPD. (5, 11, 51-53) 340 Our finding that cardiovascular disease prevalence increased with increasing severity of COPD is in 341 keeping with the body of literature on cardiovascular comorbidities and COPD, in which high 342 prevalence has been observed in (usually older) cohorts with severe airflow limitation. (5,21)  Greater polypharmacy with greater severity of COPD has also been observed previously in older 344 COPD populations, (41,54) although such analyses have been smaller (n=1859 and 398, respectively) 345 and have not assessed the specific patterns of prescribing in COPD. To the best of our knowledge, no 346 previous studies have assessed the risk of ADRs as a result of polypharmacy in COPD. A recent 347 population-based analysis of prescribing data from 310,000 adults in Scotland showed that over 15 348 years from 1995 to 2010 the proportion of people with polypharmacy and with potentially serious 349 drug-to-drug interactions increased dramatically.(35) The number of prescribed medications was 350 also associated with increased risk of interactions. Our analysis differs in approach from this analysis, 351 by seeking to identify patterns of prescribing increasing risk of specific adverse events, rather than 352 counting total potential interactions. The strength of our approach lies in highlighting specific 353 patterns of comorbidity in which specific ADRs are more likely. Our findings can therefore be applied 354 to clinical practice, highlighting the importance of recognising comorbidity in COPD and being alert 355 to specific ADRs when prescribing medication. 356

357
Our findings indicate that in those with COPD the potential for ADRs as a result of combinations of 358 medications is high, and this appears to be the result of a high prevalence of extra-pulmonary 359 comorbidities. Clinical guidelines for COPD should place greater emphasis on the need for 360 assessment of associated comorbidities and the risk of associated ADRs. While our analysis shows 361 potential areas where ADR risk exists in COPD (e.g. falls with comorbid cardiovascular disease, CNS 362 depression, constipation with comorbid mental health conditions), future research is merited to 363 assess what actual harm could be attributed to such prescribing patterns. 364

Introduction
Background/rationale 2 Explain the scientific background and rationale for the investigation being reported 4 Objectives 3 State specific objectives, including any prespecified hypotheses 5

Strengths and Limitations 53
This paper assesses multimorbidity, polypharmacy and risk of adverse drug reactions are 54 assessed in UK Biobank participants with self-reported COPD compared with those without 55

COPD. 56
Baseline variables from the UK Biobank assessment centre were used to adjust for potential 57 confounders. 58 Cumulative risk of common adverse drug reactions was quantified by identifying UK Biobank 59 participants taking three or more medications associated with similar adverse drug 60 reactions. 61 Analyses were repeated using a subgroup of participants with spirometry data confirming 62 airflow obstruction. 63 Medication and comorbidity data rely on participant self-report, and may thus be 64 susceptible to bias or inaccuracy.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  In people with Chronic Obstructive Pulmonary Disease (COPD), multimorbidity (the presence of two 69 or more long-term conditions (LTCs)) is highly prevalent.(1-4) A recent meta-analysis of 29 datasets 70 demonstrated that those with COPD are significantly more likely to be diagnosed with a range of 71 cardiovascular comorbidities than those without COPD (we will use the term comorbidity when 72 referring to specific conditions in addition to COPD, and multimorbidity to refer to the presence of 73 two or more LTCs).(5) Other LTCs with known increased prevalence in COPD include obesity,(6) 74 depression,(7-10) gastro-oesophageal reflux disease, (11)(12)(13) osteoporosis, (14)(15)(16) and lung 75 cancer. (17,18) Each of these conditions has been associated with poorer health related outcomes in 76 COPD when compared to those with no comorbidity. (19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) The overall burden of multimorbidity 77 also impacts prognosis in COPD, for example higher number of comorbidities is associated with 78 higher risk of mortality,(31) and higher burden of morbidity assessed using the Charlson index and 79 the COPD-specific comorbidity test (COTE) is associated with higher risk of all-cause and respiratory 80 specific mortality. (32,33) The importance of considering the impact of multimorbidity in the 81 management of long-term conditions is increasingly recognised, however an immature evidence 82 base means that disease specific guidelines often lack specific recommendations with respect to 83 multimorbidity.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  England, Scotland and Wales. Sociodemographic and lifestyle details were recorded using 114 touchscreen questionnaires. Townsend scores were derived from participant postcodes to provide 115 an area-based measure of socioeconomic deprivation. Self-reported LTCs, prescribed and over-the-116 counter medications, smoking status (current, previous or never) and frequency of alcohol intake 117 (never / special occasions only, one-three times a month, at least once a week) were recorded from 118 a touchscreen questionnaire and subsequent verbal interview with a study nurse. Physical activity 119 was self-reported based on a questionnaire administered in the UK Biobank assessment centre 120 http://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=6164. We classified the responses into: none (no 121 physical activity in the last four weeks), low (light 'DIY' activity only in the last four weeks), medium 122 (heavy DIY and/or walking for pleasure and/or other exercises in the last four weeks), and high 123 (strenuous sports in the last four weeks). 124 Study centre staff also collected physical measures including height and weight (to calculate body 125 mass index (BMI)) and spirometry. Spirometry was performed using a Vitalograph Pneumotrac 6800. 126 Individual reasons for contraindications to attempting spirometry were not recorded but, according 127 to protocol, these included chest infection in the last month, history of collapsed lung, and heart 128 attack or surgery in the past three months. Full details of the Biobank spirometry protocol are 129 available at https://biobank.ctsu.ox.ac.uk/crystal/docs/Spirometry.pdf. In brief, participants were 130 allowed up to three attempts to provide two reproducible spirometry measurements. Where the 131 reproducibility of the first two was deemed acceptable (<5% variation in both FEV1 and FVC) a third 132 measurement was not performed. All values were recorded along with any error messages 133 Participants reporting to have been diagnosed with chronic obstructive pulmonary disease, chronic 144 bronchitis, or emphysema at the nurse-led interview were coded as having 'self-reported COPD'. 145 Due to the potential inaccuracies of using self-reported diagnoses, we identified a subset of those 146 with self-reported COPD who met an adaptation of the Global Initiative for Obstructive Lung Disease 147 (GOLD) spirometry criteria for COPD. (46) This subset, referred to as 'GOLD COPD', was used as a 148 sensitivity analysis for self-reported COPD, and to stratify findings by severity of airflow obstruction. 149 For participants with self-report COPD and valid spirometry measurements, we calculated the ratio 150   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   9 medications with common ADRs to help identify those at cumulative risk of ADRs. This document 181 groups common medications by similar potential ADRs. While this list is not all-inclusive, and the cut-182 off value of three or more medications is arbitrary, this does allow an estimation of the cumulative 183 risk of specific ADRs. We identified six potential ADRs (falls/fractures, constipation, urinary 184 retention, CNS depression, bleeding and renal injury) for which the proportion of participants taking 185 three or more associated medications could be assessed. It should be noted that several of these 186 event (e.g. falls/fractures, CNS depression) are often multifactorial, and medication may be a 187 contributing factor rather than a definitive cause. As the guideline acknowledges, however, these 188 are clinical events of which the risk is increased by taking multiple associated medications. 189 190

Statistical analysis 191
Study hypothesis was made an analyses planned prior to inspection of the data. 192

Baseline variables 193
Comparisons were made between participants with self-reported COPD and the rest of the cohort 194 (who did not report COPD). Age, sex, smoking status, deprivation (Townsend score), BMI, physical 195 activity and frequency of alcohol intake were compared using χ 2 test for categorical variables, χ 2 test 196 for trend for ordinal variables, and Mann-Whitney-U test for continuous variables. Total number of 197 morbidities, prevalence of specific morbidities, number of self-reported prescribed medications, and 198 proportion of participants taking each class of medication (Appendix 2), were also compared 199 between those with self-reported COPD and the rest of the cohort. All comparisons were repeated 200 comparing participants with GOLD COPD only with those without COPD, stratifying by severity of 201 airflow obstruction. 202 conditions and painful conditions/inflammatory arthropathies 207 the presence of four or more morbidities (excluding COPD) 208 the use of five or more, and 10 or more, medications (two separate models) 209 Models were initially adjusted for age, sex and socioeconomic deprivation (model 1), then adjusted 210 for the addition of smoking status, alcohol frequency, BMI and physical activity (model 2). These 211 analyses were repeated comparing those with GOLD COPD only to those without COPD. 212

Risk of ADRs 213
For each potential ADR (falls/fractures, constipation, urinary retention, CNS depression, bleeding 214 and renal injury) participants taking three or more medications associated with that ADR were 215 identified. The following comparisons were then made: 216 Unadjusted percentages at risk of each ADR were calculated for participants without COPD, 217 with self-reported COPD, and with self-reported COPD plus each category of LTC 218 (cardiovascular disease, cancer, gastrointestinal disease, mental health conditions and 219 painful conditions/inflammatory arthropathies) to give an impression of the ADR risk in 220 COPD, and identify LTCs in those with COPD that may increase this risk. 221 ORs of being at risk of each ADR were calculated comparing those with self-reported COPD 222 to those without COPD adjusting for age, sex and socioeconomic deprivation (model 1) and 223 for age, sex, socioeconomic deprivation, smoking status, alcohol frequency, BMI and physical 224 activity (model 2). 225 ORs of being at risk of each ADR were calculated comparing those with and without self-226 reported COPD in each LTC category to (i.e. participants with cardiovascular disease alone 227 identify whether specific patterns of multimorbidity in COPD are associated with increased 229 ADR risk. Adjustment for a wide range of potential confounders was not appropriate in 230 these models due to the smaller number of participants in each model. 231 Each analysis was repeated comparing GOLD COPD only to those without COPD. Less than 3% of 232 participants (with or without COPD) had missing data for potential confounding variables (table 1). 233 Those with missing data were excluded from adjusted analyses. Spirometry data were missing for 234 3591 participants with self-report COPD (43%), hence the use of the GOLD COPD subset as a 235 sensitivity analysis. 236 All analyses were performed using R statistical software (version 3.     1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57 in table 2  270 comparing those with COPD, stratified by severity of airflow obstruction, with those without. Those 271 with COPD had higher numbers of LTCs and more prescribed medications than those without. There 272 was a trend towards more prescribed medications in those with greater severity of airway 273 obstruction. After controlling for age, sex and socioeconomic status, those with self-report COPD 274 were more likely to report ≥4 LTCs (3.49; 3.28 to 3.71), ≥5 medications (3.85; 3.68 to 4.03), and ≥10 275 medications (5.72; 5.36 to 6.10) than those without COPD. Results were similar for GOLD COPD and 276 remained statistically significant after adjusting for smoking status, alcohol frequency, BMI and 277 physical activity (appendix 3). 278

ADR Risk 279
Counts and percentages of participants taking specific medications are shown in appendix 4. 280 Participants with COPD (self-report and GOLD) were more likely that those without COPD to be 281 prescribed drugs across a range of disease areas, reflecting the range of LTCs present among those 282 with COPD. The percentages of participants within each category (no COPD, COPD, and COPD with 283 specific LTCs) taking three or more medications associated with a similar ADR is shown in Figure 2. 284 For each category of ADR a higher proportion of participants with COPD reported taking three or 285 more associated medications than those without COPD. This increased further with multimorbidity. 286 Participants with COPD plus cardiovascular disease had the highest percentage taking three or more 287 medications with a risk of falls or renal injury. Participants with COPD plus mental health conditions 288 had the highest percentages taking three or more medications with a risk of constipation, CNS 289 depression or bleeding. 290

Summary of main findings 312
Multimorbidity and polypharmacy in COPD were common among UK Biobank participants. The 313 presence of multimorbidity was highly prevalent in those with COPD (85%). More than half reported 314 polypharmacy (five or more medications), and 15% reported 10 or more medications. The 315 prevalence of cardiovascular disease, as well as the degree of polypharmacy, was higher among 316 those with more severe airflow obstruction. 317 For the first time, our data demonstrates that those with COPD were more likely than those without 318 to be prescribed multiple medications (≥ three) with similar ADRs. Those with COPD plus 319 cardiovascular disease were most likely to be taking multiple medications associated with increased 320 risk of falls or renal injury, while those with COPD plus mental health conditions were most likely to 321 be taking medications predisposing to constipation, CNS depression and bleeding.(50) Within each 322 category of LTC, those with COPD were more likely to be taking multiple medications with similar 323 ADRs than those without. These associations between patterns of multimorbidity and specific ADR 324 risks have not been described or quantified previously. 325 326

Strengths and limitations 327
Strengths of this study include the large sample size with representation from different areas of the 328 UK. The range of data collected at UK Biobank assessment centres meant it was possible to compare 329 a range of sociodemographic characteristics as well as spirometry data, the latter being unusual for a 330 large community based cohort. It is recognised, however, that UK Biobank participants show some 331 evidence of 'healthy volunteer bias', differing from the UK average on a number of socioeconomic, 332 lifestyle and health-related measures. Specifically they are less socioeconomically deprived, less 333 likely to smoke, to be obese, and have fewer self-reported health conditions.(51) All LTC diagnoses 334  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   19 as well as medication data were self-reported, with no alternative means of verification. We 335 attempted to minimise this limitation by identifying a subset of those with COPD meeting the GOLD 336 diagnostic criteria and repeating the analyses with this subset. Importantly, spirometry values were 337 also pre-bronchodilator, which is in contravention to guidelines for diagnosing COPD. Additionally, 338 information was not available about the strength of indication for medications and individual 339 susceptibility to risk, which is a limitation when considering the risk of ADRs. 340 The use of the Scottish Government Polypharmacy Guideline allowed analysis of potential ADR risk 341 by specific common ADRs. The intended purpose of this guideline, however, was not to identify 342 potential risk from a population sample, but rather to identify potential causes of symptoms or 343 complications. The analysis in this study, therefore, serves only as an approximation of potential risk, 344 not an absolute marker of inappropriate polypharmacy. The cross-sectional nature of this analysis 345 also precludes an analysis of actual harm as a result of polypharmacy. Many of the potential ADRs, 346 such as falls and fractures and renal injury, and frequently multifactorial events and may not be 347 directly attributable to medication use. Despite these limitations, however, the co-prescription of 348 multiple medications with similar ADRs strongly implies greater potential for harm. The association 349 of such prescribing patterns with COPD, across a range of potential ADRs, is clear from our findings. 350 This analysis is, to the author's knowledge, the first to attempt to quantify this risk for specific ADRs 351 in this way. 352 353

Conclusions:
Multimorbidity is common in COPD and associated with high levels of polypharmacy.

Strengths and Limitations
This paper assesses multimorbidity, polypharmacy and risk of adverse drug reactions in UK Biobank participants with self-reported COPD compared with those without COPD.
Baseline variables from the UK Biobank assessment centre were used to adjust for potential confounders.
Cumulative risk of common adverse drug reactions was quantified by identifying UK Biobank participants taking three or more medications associated with similar adverse drug reactions.
Analyses were repeated using a subgroup of participants with spirometry data confirming airflow obstruction.

Defining COPD
Participants reporting to have been diagnosed with chronic obstructive pulmonary disease, chronic bronchitis, or emphysema at the nurse-led interview were coded as having 'self-reported COPD'.
Due to the potential inaccuracies of using self-reported diagnoses, we identified a subset of those with self-reported COPD who met an adaptation of the Global Initiative for Obstructive Lung Disease (GOLD) spirometry criteria for COPD. (48) This subset, referred to as 'GOLD COPD', was used as a sensitivity analysis for self-reported COPD, and to stratify findings by severity of airflow obstruction.
For participants with self-report COPD and valid spirometry measurements, we calculated the ratio of forced expiratory volume in 1 second (FEV1) to forced vital capacity (FVC) using the highest   off value of three or more medications is arbitrary, this does allow an estimation of the cumulative risk of specific ADRs. We identified six potential ADRs (falls/fractures, constipation, urinary retention, CNS depression, bleeding and renal injury) for which the proportion of participants taking three or more associated medications could be assessed. It should be noted that several of these event (e.g. falls/fractures, CNS depression) are often multifactorial, and medication may be a contributing factor rather than a definitive cause. As the guideline acknowledges, however, these are clinical events of which the risk is increased by taking multiple associated medications.

Statistical analysis
All analyses were planned prior to inspection of the data.

Baseline variables
Comparisons were made between participants with self-reported COPD and the rest of the cohort (who did not report COPD). Age, sex, smoking status, deprivation (Townsend score), BMI, physical activity and frequency of alcohol intake were compared using χ 2 test for categorical variables, χ 2 test for trend for ordinal variables, and Mann-Whitney-U test for continuous variables. Total number of morbidities, prevalence of specific morbidities, number of self-reported prescribed medications, and proportion of participants taking each class of medication (Appendix 2), were also compared between those with self-reported COPD and the rest of the cohort. All comparisons were repeated comparing participants with GOLD COPD only with those without COPD, stratifying by severity of airflow obstruction.

Multimorbidity and polypharmacy
Logistic regression analyses were used to compare participants with self-reported COPD and those without COPD. Odds ratios (OR) and 95% confidence intervals (95% CI) were calculated for: the use of five or more, and 10 or more, medications (two separate models) Models were initially adjusted for age, sex and socioeconomic deprivation (model 1), then adjusted for the addition of smoking status, alcohol frequency, BMI and physical activity (model 2). These analyses were repeated comparing those with GOLD COPD only to those without COPD.

Risk of ADRs
For each potential ADR (falls/fractures, constipation, urinary retention, CNS depression, bleeding and renal injury) participants taking three or more medications associated with that ADR were identified. The following comparisons were then made: Unadjusted percentages at risk of each ADR were calculated for participants without COPD, with self-reported COPD, and with self-reported COPD plus each category of LTC (cardiovascular disease, cancer, gastrointestinal disease, mental health conditions and painful conditions/inflammatory arthropathies) to give an impression of the ADR risk in COPD, and identify LTCs in those with COPD that may increase this risk.
ORs of being at risk of each ADR were calculated comparing those with self-reported COPD to those without COPD adjusting for age, sex and socioeconomic deprivation (model 1) and for age, sex, socioeconomic deprivation, smoking status, alcohol frequency, BMI and physical activity (model 2).
ORs of being at risk of each ADR were calculated comparing those with and without selfreported COPD in each LTC category to (i.e. participants with cardiovascular disease alone compared to participants with cardiovascular disease plus COPD, etc.). This was intended to identify whether specific patterns of multimorbidity in COPD are associated with increased Each analysis was repeated comparing GOLD COPD only to those without COPD. Less than 3% of participants (with or without COPD) had missing data for potential confounding variables (table 1).
Those with missing data were excluded from adjusted analyses. Spirometry data were missing for 3591 participants with self-report COPD (43%), hence the use of the GOLD COPD subset as a sensitivity analysis.
All analyses were performed using R statistical software (version 3.3.1).  Figure 1) and are referred to here as GOLD COPD. Table 1 describes and compares the characteristics of those with and without COPD in UK Biobank.

Baseline variables
Participants with COPD (both self-report and GOLD) were significantly older, more socioeconomically deprived, and less physically active. A higher proportion of those with COPD were male, obese and had a history of smoking.

Multimorbidity and polypharmacy
Prevalence of each category of comorbidity was higher in those with COPD than without (   in table 2 comparing those with COPD, stratified by severity of airflow obstruction, with those without. Those with COPD had higher numbers of LTCs and more prescribed medications than those without. There was a trend towards more prescribed medications in those with greater severity of airway obstruction. After controlling for age, sex and socioeconomic status, those with self-report COPD were more likely to report ≥4 LTCs (3.49; 3.28 to 3.71), ≥5 medications (3.85; 3.68 to 4.03), and ≥10 medications (5.72; 5.36 to 6.10) than those without COPD. Results were similar for GOLD COPD and remained statistically significant after adjusting for smoking status, alcohol frequency, BMI and physical activity (appendix 3).

ADR Risk
Counts and percentages of participants taking specific medications are shown in appendix 4.
Participants with COPD (self-report and GOLD) were more likely that those without COPD to be prescribed drugs across a range of disease areas, reflecting the range of LTCs present among those with COPD. The percentages of participants within each category (no COPD, COPD, and COPD with specific LTCs) taking three or more medications associated with a similar ADR is shown in Figure 2.
For each category of ADR a higher proportion of participants with COPD reported taking three or more associated medications than those without COPD. This increased further with multimorbidity.
Participants with COPD plus cardiovascular disease had the highest percentage taking three or more medications with a risk of falls or renal injury. Participants with COPD plus mental health conditions had the highest percentages taking three or more medications with a risk of constipation, CNS depression or bleeding.
After adjusting for age, sex and socioeconomic deprivation, those with self-report COPD remained more likely to be taking three or more medications in each category than those without COPD. These  (Table 3). Findings were similar for GOLD COPD however, after adjusting for additional potentially confounding variables, results for bleeding risk were not statistically significant in this sensitivity analysis (Table 3). comparing those with and without COPD (e.g. participants with cardiovascular disease plus COPD compared with participants with cardiovascular disease alone, etc.). These models were adjusted for age, sex and socioeconomic status only. Within each category of LTC, those with self-reported COPD were more likely to be at risk of each ADR than those without COPD (appendix 3). Not all results were statistically significant when using GOLD COPD (Appendix 3).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 18

Summary of main findings
Multimorbidity and polypharmacy in COPD were common among UK Biobank participants. The presence of multimorbidity was highly prevalent in those with COPD (85%). More than half (52%) reported polypharmacy (five or more medications), and 15% reported 10 or more medications. The prevalence of cardiovascular disease, as well as the degree of polypharmacy, was higher among those with more severe airflow obstruction.
For the first time, our data demonstrates that those with COPD were more likely than those without to be prescribed multiple medications (≥ three) with similar ADRs. Those with COPD plus cardiovascular disease were most likely to be taking multiple medications associated with increased risk of falls or renal injury, while those with COPD plus mental health conditions were most likely to be taking medications predisposing to constipation, CNS depression and bleeding.(52) Within each category of LTC, those with COPD were more likely to be taking multiple medications with similar ADRs than those without. These associations between patterns of multimorbidity and specific ADR risks have not been described or quantified previously.

Strengths and limitations
Strengths of this study include the large sample size with representation from different areas of the UK. The range of data collected at UK Biobank assessment centres meant it was possible to compare a range of sociodemographic characteristics as well as spirometry data, the latter being unusual for a large community based cohort. It is recognised, however, that UK Biobank participants show some evidence of 'healthy volunteer bias', differing from the UK average on a number of socioeconomic, lifestyle and health-related measures. Specifically they are less socioeconomically deprived, less likely to smoke, to be obese, and have fewer self-reported health conditions.(53) All LTC diagnoses  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   19 as well as medication data were self-reported, with no alternative means of verification. We attempted to minimise this limitation by identifying a subset of those with COPD meeting the GOLD diagnostic criteria and repeating the analyses with this subset. Importantly, spirometry values were also pre-bronchodilator, which is in contravention to guidelines for diagnosing COPD. Additionally, information was not available about the strength of indication for medications and individual susceptibility to risk, which is a limitation when considering the risk of ADRs.
The use of the Scottish Government Polypharmacy Guideline allowed analysis of potential ADR risk by specific common ADRs. The intended purpose of this guideline, however, was not to identify potential risk from a population sample, but rather to identify potential causes of symptoms or complications. The analysis in this study, therefore, serves only as an approximation of potential risk, not an absolute marker of inappropriate polypharmacy. The cross-sectional nature of this analysis also precludes an analysis of actual harm as a result of polypharmacy. Many of the potential ADRs, such as falls and fractures and renal injury, and frequently multifactorial events and may not be directly attributable to medication use. Despite these limitations, however, the co-prescription of multiple medications with similar ADRs strongly implies greater potential for harm. The association of such prescribing patterns with COPD, across a range of potential ADRs, is clear from our findings. This analysis is, to the author's knowledge, the first to attempt to quantify this risk for specific ADRs in this way.

Context and implications
The increased prevalence of individual LTCs such as coronary heart disease, hypertension, diabetes, dyspepsia, osteoporosis, cancer, depression and anxiety in those with COPD is similar to the findings from other population based studies of multimorbidity in COPD. (5,11,(54)(55)(56) Our finding that cardiovascular disease prevalence increased with increasing severity of COPD is in keeping with the  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 37) The number of prescribed medications was also associated with increased risk of interactions. Our analysis differs in approach from this analysis, by seeking to identify patterns of prescribing increasing risk of specific adverse events, rather than counting total potential interactions. The strength of our approach lies in highlighting specific patterns of multimorbidity in which specific ADRs are more likely. Our findings can therefore be applied to clinical practice, highlighting the importance of recognising multimorbidity in COPD and being alert to specific ADRs when prescribing medication.
Our findings indicate that in those with COPD the potential for ADRs as a result of combinations of medications is high, and this appears to be the result of a high prevalence of extra-pulmonary LTCs.
Clinical guidelines for COPD should place greater emphasis on the need for assessment of associated multimorbidity and the risk of associated ADRs. While our analysis shows potential areas where ADR risk exists in COPD (e.g. falls in those with concomitant cardiovascular disease, CNS depression, constipation with concomitant mental health conditions), future research is merited to assess what actual harm could be attributed to such prescribing patterns.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   21 Among UK Biobank participants with COPD there was considerable multimorbidity and polypharmacy. Those with COPD were highly likely to be concurrently prescribed multiple medications with similar potential adverse effects. Medications contributing to this risk were largely indicated for the management of the associated morbidities rather than COPD. Future research should examine the effects on healthcare outcomes of co-prescribing of multiple drugs with similar potential of ADRs. Clinical guidelines for COPD should emphasise the need for assessment of multimorbidity and the risk of associated ADRs.

Introduction
Background/rationale 2 Explain the scientific background and rationale for the investigation being reported 4 Objectives 3 State specific objectives, including any prespecified hypotheses 5