Introduction  

Osteoporosis in both male and female patients with chronic obstructive pulmonary disease (COPD) is firmly established as one of the core comorbid conditions.1 2 Over the last decade, it has become clear that osteoporosis is not just an end-stage COPD problem3 nor just in those on maintenance oral corticosteroids (OCS), but it also occurs in a large proportion of those with mild–moderate airflow obstruction and even in steroid naïve patients.4 5 The Global Initiative for COPD (GOLD) strategy recommends that osteoporosis co-existence should be considered in COPD,1 and that the UK National Institute for Health and Care Excellence (NICE) Guidelines on osteoporosis considers COPD as a secondary cause of osteoporosis encouraging the use of fracture prediction tools.6

The causes for osteoporosis in COPD are likely multiple and cumulative, including age, smoking exposure, inactivity, low body mass index (BMI), systemic inflammation and the frequent use of OCS.7 The clinical implications of osteoporosis include increased risk of fractures, poor quality of life, pain and further deterioration in lung function.8 9 Osteoporosis can also remain undiagnosed as asymptomatic for many years.10 Fractures are a function of trauma sustained, such as falls which are common in COPD,11 and affect the quality and architecture of the bone. Fractures contribute further pain, poor quality of life, increased mortality and confer a substantial economic burden on health systems, patients and their families.12 13 Given this, the individual risk of a future fracture in patients with COPD is crucial to determine in-patient care and to treat accordingly.

Fracture risk prediction tools based on clinical and personal characteristics have been developed over the years to guide the investigation and management of those identified to be at high risk of osteoporotic fractures, worldwide. These include for the UK (and many other regions), FRAX and QFracture.6

The full extent of fracture risk assessment in patients with COPD is not fully established. The aim of this study was to assess the incidence of hip fracture alone or all major osteoporotic fractures (MOF) in patients with COPD compared with non-COPD patients, and to evaluate the use and performance of fracture risk prediction tools in patients. Further, to assess the prevalence of coded osteoporosis up to the time of COPD diagnosis and the incidence of osteoporosis.

Methods

Information for this cohort study was obtained from The Health Improvement Network (THIN), an anonymised primary care database representing 6.2% of the total UK population.14

The study population consisted of patients 40 years and over with a new READ-coded COPD diagnosis during the data collection period 1st January 2004 to 31st December 2015, with at least 1 year of record prior to COPD diagnosis.15 Each patient was matched by age, sex and general practice (GP) to up to four patients without a history of COPD to generate a matched cohort and assigned the same index date.

Follow-up was from the index date to the first record of either the occurrence of the outcome of interest (fracture/osteoporosis), the date of transfer of the patient out of the practice area, death or the end of THIN data collection. Diagnoses for osteoporosis—classed as coded osteoporosis (online appendix 1), hip fracture alone and all MOF (comprising fracture of the hip, proximal humerus, forearm or clinically symptomatic vertebra/spine), coded using the standard READ code classification were used.16

A series of explanatory variables6 17 determined at baseline (prior to or at index date) included Charlson Comorbidity Index (CCI) score,18 Townsend social deprivation score, recorded history of fall, prior fractures, parental history of fall/osteoporosis, relevant comorbidities and secondary causes of osteoporosis as defined in the FRAX questionnaire.19 Records for smoking status, alcohol use, Medical Research Council (MRC) Dyspnoea scale, BMI and use of specific prescription drugs were restricted to a defined time period.

To account for the use of OCS, individual follow-up time was divided into periods during which participants were considered exposed, or not exposed, to OCS (a binary variable). Exposed periods started from prescription date until the first gap of more than 90 days between prescriptions, with individuals considered unexposed from the 91st day onwards. Individuals were considered exposed at study entry if they had received a relevant prescription within 90 days prior. The effect of exposure was assumed to be constant, and not cumulative, over time.

Input variables included clinical status, prescription drug use, and demographic characteristics, according to the variables/definitions used in both FRAX and QFracture tools,19 20 additional detail on the method is provided in an online data supplement (online appendix 2). Imputation was used for missing variables.

The 10-year risk score for hip fracture alone and all MOF according to QFracture (V.2017.0.0.0) and FRAX for UK (without bone mineral density [BMD] information) (desktop V.3.12) were calculated for patients with COPD, aged 40–90 years. A complete case sensitivity analysis without imputed variables was also performed (online appendix 3).

Results

The baseline characteristics are shown in table 1. A total of 80 874 eligible patients with COPD and 308 999 matched non-COPD patients were identified. The median follow-up period was 5 years for both patients with COPD and non-COPD patients.

Fracture risk prediction tools in COPD

Only 1074 (1.33%) of the patients with COPD had a FRAX assessment READ-coded ever documented in the records and 12 patients had a READ coded QFracture assessment. Within 1 year (before and after) of index date, 248 (0.31%) of the patients with COPD had a FRAX and only one patient a QFracture.

The final population for the discriminatory accuracy analysis comprised 72 559 patients aged 40–90 years with COPD and no prior diagnosis of osteoporosis or prescription of any anti-resorptive treatment (demographics in onlineappendix 6). This included 4605 (6.4%) patients who experienced any MOF and 1444 (2.0%) who experienced a hip fracture.

When the FRAX and QFracture scores were calculated for patients with COPD, for hip fracture there were 29 035 (40.0%) patients who had a risk≥3% using FRAX and 33 065 (45.6%) patients using QFracture. For any MOF, 6221 (8.6%) of the patients had a risk≥20% using FRAX and 9546 (13.2%) patients using QFracture.

Both risk tools had a similar discriminatory accuracy for hip fracture (FRAX 76.1%, 95% CI 74.9% to 77.2% and QFracture 76.1%, 95% CI 74.9% to 77.2%). FRAX, however, had a higher accuracy for MOF (71.4%–95% CI 70.6% to 72.2%) than QFracture (61.4%–95% CI 60.5% to 62.3%). The discriminatory accuracies were better in women than men. The performance of the prediction tools was similar in patients aged 50–90 years compared with those aged 40–90 years.

The sensitivity of the risk scores for any MOF (using >20% risk as cut-off) was similar: FRAX: 25.4% and QFracture: 25.2%. The sensitivity of the risk scores for hip fracture (using >3% cut-off) was slightly worse for FRAX: 78.1% compared with 82.1% for QFracture. The specificity and positive predictive value were better for FRAX than QFracture, table 3.

View this table:

• View inline
• View popup

Table 3

Discrimination measures for FRAX and QFracture at recommended treatment cut-offs for both MOF and hip fractures alone

The association of an increased fracture risk (either FRAX or QFracture) with the incidence of any MOF is shown in figure 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Kaplan-Meier plots comparing the incidence of MOF at various predicted fracture risk categories in patients with COPD using (A) FRAX and (B) QFracture. COPD, chronic obstructive pulmonary disease; MOF, major osteoporotic fractures.

Discussion

Using UK primary care electronic health records, we have reported on the burden of fractures in patients with COPD with both hip fracture alone or any MOF increased in patients with COPD compared with age, sex and GP surgery matched patients. Despite the increased fracture risk and recommendations in the NICE osteoporosis guidelines, fracture risk prediction tools are rarely coded. However, where the risk score was retrospectively calculated, the risk prediction tools identify those at risk of hip fracture or any MOF. Therefore, fracture risk prediction and subsequent targeted therapy and management could transform multi-morbidity management of COPD. In addition, we report that the prevalence and incidence of osteoporosis, a risk for fracture, in patients with COPD, is far greater than in non-COPD patients.

Prevalence of osteoporosis varies widely in the different research studies of patients with COPD. This is likely due to the severity of COPD,4 5 whether osteoporosis was systematically sought or self-reported,4 22 and whether patients included were on OCS.3 A prevalence of 23%–32% has been reported where BMD was systematically performed in COPD,4 23 while 14% of the patients with COPD self-reported osteoporosis compared with 5% in those without COPD.22 The prevalence of coded osteoporosis in the GP health records presented here was, however, far lower at 5.7% than the reported prevalence from clinical studies when osteoporosis and BMD are systematically assessed. This raises the question of subclinical, undiagnosed osteoporosis leading to a missed opportunity for intervention and strengthening the need for a systematic assessment, especially when cost-efficient anti-resorptive treatment is available.24

There is growing consensus on COPD being a secondary cause of osteoporosis, including within the NICE clinical guidelines on osteoporosis where fracture risk prediction tools are recommended, yet in practice seem rarely done.6While osteoporosis in itself leads to pain and poor quality of life,25 ultimately osteoporosis treatment aims to reduce the risk of fracture.24 26 Risk factors for fracture include osteoporosis but also falls, which, are greater in patients with COPD.11 27 While the increased risk of fractures in COPD has previously been considered,28 they have not assessed the incidence from the time of COPD diagnosis or only reported as part of a larger study of post-menopausal women29 or analysed the history of obstructive airway disease (COPD and asthma together) before the index date of osteoporotic fracture in both cases and controls over the age of 18 years.30

Little is known about the use of fracture risk assessment tools in patients with COPD. A number of validation studies have performed independent assessments to predict subsequent fracture in the general population.31 32 The studies differ widely in sample size, methodology, and techniques used to assess performance.33 Discrimination for FRAX (without BMD incorporation) and QFracture have both been reported as good.31 34 35 The results from this COPD study are comparable to the general population validation studies;31 34 35 however, the area under curve for MOF using QFracture was lower than that reported in other studies. Similar to findings from studies based on general population, the discrimination from our study was better in women than men and better for hip fracture than MOF.36 The discrimination appeared similar within the 50–90-year-old group when compared with the 40–90-year olds. Despite the two tools having differences in their approach to calculating fracture risks, both predict fractures satisfactorily in patients with COPD. Despite the sensitivity and positive predictive values being far from ideal, sensitivity reported in our study are comparable to those published in studies using a general population.31 35 Although a bespoke COPD tool could be adapted in the future, the use of one of the established fracture risk scores in the meantime provides the opportunity to systematically identify and intervene. Such tools are incorporated into primary care medical record systems and utilised in a number of other disease areas. The available fracture prevention therapy (anti-resorptive agents) are very effective, safely yielding 40%–60% reduction in the risk of fracture.26 These medications are cost-effective in high-risk patients—reduces morbidity, mortality and the healthcare cost associated with osteoporotic fractures.24 The fracture prediction tools could be integrated into COPD annual assessments or at COPD diagnosis. The identification of patients at high risk is a valuable information to guide and optimise treatment options, though the optimal pathways for this integration is required.

The use of OCS has been considered to be a major contributory factor in the development of osteoporosis. However, osteoporosis has been reported in patients with no OCS use.4 5 Other known osteoporosis risk factors are also likely to contribute in patients with COPD, including smoking, a low BMI, physical inactivity and systemic inflammation. Some of these risk factors could be moderated through education, smoking cessation, pulmonary rehabilitation and lifestyle changes.37 38 Recognition of the scale and impact of fracture risk draws further necessary attention to these interventions to aim to prevent and reduce risks, alongside appropriate pharmacotherapy.

The study had several strengths in its methods, analyses, findings and implications for clinical practice. First, this research was population-based and compared patients with COPD with age-sex-matched control patients from the same GP. Its external validity and hence generalisability was high because the THIN database is representative of the UK population.14 There was a substantial duration of follow-up. A wide range of potential confounders were also evaluated and adjusted for in the analyses.

For the assessment of the fracture prediction tools, the population of patients with COPD used was large, with many fracture events, and included both men and women. This minimised the likelihood of a selection bias. The assessments of the prediction tools were done using the same population, therefore minimising the effect of confounding for a difference in performance. We are presently not aware of studies that have determined the performance of the recommended fracture prediction tools in the sub-population of patients with COPD. The dataset was using UK electronic health records but is likely representative of other countries in representing the scale of the problem and the utility of the risk prediction scores.

Regarding limitations, some variables might be subject to information or reporting bias as READ codes recorded in databases do not capture free text from consultations. Such variables include patient-reported alcohol intake, use of cigarettes or their awareness of relevant family history. The possibility of residual confounding can also not be excluded as risk factors such as physical activity, diet and ethnicity could not be adjusted for in the analyses. An accepted definition of fracture types was used; however, it is difficult to determine the cause of fracture based simply on the fracture site, with no additional information. Unlike studies which assess BMD systematically, this is not currently done in clinical practice, nor are the fracture risk scores routinely calculated as highlighted here. Therefore, the incidence of osteoporosis based on clinical codes likely reflects an underestimation of the true increased incidence/risk of osteoporosis.

In summary, despite validated fracture risk prediction tools, there was very little assessment of the increased fracture risk in patients with COPD. However, on retrospective calculation of fracture risk, the tools identify those patients with COPD at greatest risk of fracture. The identification with a systematic assessment of bone health and addressing prevention and treatment of those at a greater risk of fracture have the potential to improve outcomes for patients with COPD.