Identifying Dysphagia and Demographic Associations in Older Adults Using Electronic Health Records: A National Longitudinal Observational Study in Wales (United Kingdom) 2008–2018

Older people are a becoming a larger percentage of the world’s population, with 15% of the UK population aged 65 years and over [1, 2]. In the UK demography, the 85 + age group (very old) is the fastest growing cohort and is estimated to double to 3.2 million by mid-2041 and treble by 2066 (5.1 million; 7% of the UK population) [3‐6]. More concerning is that over the last 15 years, there has been an 85% increase in the number of centenarians, which is predicted to reach 21,000 by 2030 [6].

Swallowing is a complex interplay of the cerebral cortex and brainstem, several cranial nerves and multiple muscles in the head and neck [7]. The main purpose of swallowing is to transfer food safely from the mouth to the stomach whilst protecting the airway. There are three sequential phases of swallowing: oral, pharyngeal and oesophageal. The pharyngeal swallow is semi-automatic and occurs in a coordinated sequence, the timings of which will be modulated by the cortex depending on the characteristics of the bolus/food swallowed [8]. Dysfunction (secondary to localised or generalised disease processes) in one or more of the phases may result in dysphagia (difficulty swallowing) [9, 10]. Dysphagia is defined as a subjective sensation of difficulty or abnormality of swallowing [11]. This definition encompasses oropharyngeal dysphagia, oesophageal dysphagia and reflux disease.

Changes to eating and swallowing may occur with age, with some individuals accepting this as a normal part of ageing, slowly adapting the quantity and texture of food eaten and the time taken to eat, often referred to as presbyphagia [11, 12]. Physiological changes to the swallow with increasing age may vary with gender and between individuals. In general, it has been found that there is reduced hyoid excursion, laryngeal elevation and anterior rotation in individuals with dysphagia. Transit times through the pharynx increase, and UES opening is prolonged [13]. When does a change in swallow pattern become pathological? Humbert and Robbins [14] cautioned against labelling a “normal yet safe swallow” as being evidence of dysphagia. McCullough et al. comment that laryngeal penetration was more frequent and deeper in the older age group, but that clearance was the usual outcome [15]. Presbyphagia [11] is defined as “characteristic changes in the mechanism of oropharyngeal/oesophageal swallowing of otherwise healthy older adults” which may manifest as fatigue [16] and slowness of eating rather than overt dysphagia. The identified physiological changes with age [17, 18] may be directly or indirectly secondary to frailty and pre-existing comorbidities, rather than age specific.

There have been many prevalence/epidemiological studies of dysphagia in community living older people [15, 19‐26]. These surveys have been completed over three decades and have used several different methodologies including postal surveys, cold-calling telephone calls and clinic waiting room approaches. The studies have also used variable definitions of “old” and of what constitutes dysphagia [27], which makes comparisons difficult.

Determining/understanding the prevalence of dysphagia in the community is complex. Studies are referenced to a particular country, set environment or disease state, rather than an unselected population. Often the definition of dysphagia is vague or not stated. The true prevalence of dysphagia in the community is uncertain, with a wide range from 2.5 to 72% [17, 25, 27‐32] reported. Adkins et al. reported a prevalence of 16.1% (4998/31129) across all age groups [20], whereas Battacharyya reported a 4% prevalence from a cohort of 30,000 in a household survey [33]. Kertscher et al. using EAT-10 as part of a telephone survey, found that 21.9% of those > 76 years scored ≥ 3, indicating problems swallowing [21]. A study by Andersen et al. found that 26% of participants reported difficulty taking tablets [34]. In two studies of community living older (> 65 years) adults, at different time points [35, 36], 33.7% of those in Korea were found to have symptoms of dysphagia, whilst the prevalence was only 13.8% in Japan. Patel et al. reported that 3.1% of all adult admissions to hospital had a diagnosis of dysphagia [37].

Frailty is a modern geriatric syndrome (giant) [38], but identifying frailty can be difficult due to the lack of a universally accepted definition [18]. Frailty has been defined by some as “a biologic disorder affecting many activities of daily living characterised by diminishing physiological reserves [39, 40] which may be due to an accumulation of multiple system deficits” [41], and as a “state of reduced strength and physiological malfunctioning that increases a person’s susceptibility to increased dependency, vulnerability and even death” [42, 43]. It is generally accepted that the prevalence of frailty increases with age, with approximately 25–50% of people over the age of 85 years being frail [44]. In hospital populations, studies have shown that frailty and dysphagia frequently co-exist [14, 45, 46] but there remains a debate as to whether there is a distinct entity of frailty dysphagia [14]. Recent papers [38, 47] define oral frailty—but not frailty dysphagia—as a decrease in oral functioning together with a decline in cognitive and physical function but fail to fully develop the discussion. Tanaka et al. noted that poor oral status in ≥ 1 oral domain was a predictor for physical frailty, suggesting association rather than a distinct entity of oral frailty. However, it is to be expected that dysphagia co-exists with frailty due to the suprahyoid muscle complex and the masseter muscle being skeletal in nature [48].

Frail older adults admitted to hospital have a prevalence of dysphagia of 55% [14, 45]. Cha et al. report that sarcopenia and dysphagia are concurrent in 6% of community dwelling older adults (> 65 years) [49]. Hansen et al. found that at a follow-up assessment of 56 weeks after discharge from hospital, 25% of older adults had dysphagia and that there was a significant relationship at the p < 0.05 level with sarcopenia, low physical performance and malnutrition [50]. These changes, combined with changes to food preferences, have the potential to result in under-nutrition [18]. These inter-relationships contribute to a “failure to thrive” and have the potential to accelerate people towards a terminal physical decline ultimately resulting in death.

Nursing home populations have a greater prevalence due to many residents having a diagnosis of dementia, stroke or Parkinson’s disease alone, or in combination, with frailty [51, 52]. In a systematic review by Takizawa et al., dysphagia was diagnosed in 8.1–80% of stroke patients, 27–30% of brain injury patients, 11–81% of Parkinson disease patients and 91.7% of older persons with community-acquired pneumonia due to impaired swallowing and breathing mechanisms [53].

In the clinical situation, despite clinical staff being aware that dysphagia is a complication of many aetiological mechanisms, including mechanical, neurological and psychological [54], it is nevertheless often under-diagnosed and under-treated [55]; as a consequence, there is an increased risk of malnutrition, dehydration, pneumonia [56] and mortality [57] in the relevant populations.

Studies may over or under report the prevalence of dysphagia in the community depending on the research approach taken. Bias may be introduced by both the survey and the researcher. Those studies of dysphagia focussing on older people have tended to be small [17, 26], and larger studies have relied on census data [24, 33]. No studies have examined the medical data held in electronic health records (EHRs) that have been collated as part of day-to-day clinical care and explored the relationship with and age, gender, frailty and deprivation.

The use of existing anonymised routinely collected longitudinal data can help provide rapid access to large-scale data for studies and provide robust evidence for commissioning decisions and policy [58, 59]. In this study, we utilise the Secure Anonymised Information Linkage (SAIL) Databank [60‐62] to investigate the recording of dysphagia in EHRs in older adults in Wales over time. We also include linked data to explore associations between dysphagia and age, gender, frailty and deprivation.

We used anonymised EHRs and administrative data from the SAIL Databank to create a longitudinal cross-sectional cohort study.

For the purposes of this study, we have accepted dysphagia as any entry in to the EHR that documents dysphagia or swallowing problem/difficulty.

Our cohorts were created using data held within the SAIL Databank [60‐62]. The SAIL Databank contains longitudinal anonymised administrative and healthcare records for the population of Wales. The anonymisation is performed by a trusted third party, the National Health Service Wales Informatics Service. The SAIL Databank has a unique individual anonymised person identifier known as an anonymous linking field and a unique address anonymised identifier known as a residential anonymous linking field [63] that are used to link between data sources at individual and residential levels, respectively. Individual linking fields, nested within residential codes, are contained in the anonymised version of the Welsh Demographic Service Dataset (WDSD), replacing the identifiable names and addresses of people registered with a free-to-use General Practitioner service.

The WDSD contains demographic information including week of birth, gender, date of death and lower-layer super output area (LSOA). Each LSOA in Wales has been ranked from most deprived to least deprived according to the Welsh Index of Multiple Deprivation (WIMD), an area-based measure of socio-economic status [64]. We used the 2011 version of LSOA linked to the 2014 version of the WIMD and used deprivation quintiles in the analyses. To detect dysphagia in older adults, we used both the Welsh Longitudinal General Practice (WLGP) dataset and the Patient Episode Database for Wales (PEDW). The WLGP contains general practice data for approximately 80% of the population of Wales, and we used Read codes version 2 (CTV2) to identify dysphagia. The PEDW dataset contains all hospital admission records in Wales, and we identified individuals with dysphagia in these records using the international classification of disease version 10 (ICD-10) code for dysphagia.

We included all individuals aged 65 + living in Wales with a valid address who were registered with a general practice contributing data to the SAIL databank. We created 11 cohorts independently, one for each year from 2008 to 2018. Dysphagia was identified in the previous year from the cohort index dates to mitigate the need for censoring due to death or migration. For example, the cohort created for 2008 had index date 1st January 2008, and dysphagia was identified from 1st January 2007 to 31st December 2007.

Our outcome of interest was dysphagia identified in either general practice or hospital records. We used CTV2 Read codes to identify dysphagia in general practice records and ICD-10 codes to identify dysphagia in hospital records. We created binary indicators for dysphagia for each person, each year if dysphagia was identified in either general practice, hospital records, or both in the year prior to the cohort index date. No restriction was placed on the number of codes searched in either general practice or hospital records. The cohort index dates were treated as categorical and included as the year of the cohort index date (2008 to 2018). Age was calculated on the cohort index date and was categorised in to three groups: 65–74, 75–84, 85+. Gender (male/female) was treated as categorical variable. Frailty was calculated using the electronic Frailty index (eFI). The eFI is based on an internationally established cumulative deficit model and assigns a frailty score to an individual calculated using 36 variables from primary care GP data, including symptoms, signs, diseases, disabilities and abnormal laboratory values, referred to as deficits [65]. The eFI score is the number of deficits present, expressed as an equally weighted proportion of the total. An individual with a single deficit would be assigned an eFI of 1/36 (0.03); another with nine deficits would be assigned an eFI of 9/36 (0.25). The eFI score is used to categorise individuals as fit (eFI value of 0–0.12), mild (> 0.12–0.24), moderate (> 0.24–0.36), or severely frail (> 0.36) [65]. We calculated the eFI retrospectively in the SAIL databank on the cohort index dates using 10 years of previous GP data for each individual each year [66]. This meant that for someone in the cohort for the 1st January 2010, the eFI was calculated using data from 1st January 2000 to 1st January 2010. The WIMD quintiles were included as categorical variables ranging from 1—Most Deprived, to 5—Least Deprived.

We produced descriptive statistics for our cohorts and stratified these by individuals recorded as having dysphagia and those not. We computed chi-squared statistics with a significance level of 0.05 for each cohort year to investigate differences in age, frailty, gender and WIMD between individuals with and without a dysphagia diagnosis. We calculated multilevel logistic regression models to investigate associations between dysphagia recorded in the previous year and the demographic information. The multilevel logistic regression models included the cohort year as a random effect. Sensitivity analyses included independent multilevel logistic regression models for dysphagia diagnoses in primary care (general practice) and secondary care (hospital) records. Statistical analyses were performed using R version 4.0.0 [67] with R2MLwiN [68].

We developed our cohorts by limiting to only individuals aged 65 + on the index dates, this was further limited by restricting the cohort to individuals registered with a general practice submitting data to the SAIL databank. Finally, we restricted the cohorts to individuals with a valid residential address so we were able to determine the WIMD for each individual. The remaining number of people after these restrictions is recorded in the supplementary material, Table S1.

The number of people diagnosed with dysphagia varied with each year. We further stratified the diagnoses by Read code and ICD-10 code. The numbers of codes are recorded in the supplementary material, Table S2. The results suggest an increase in the recording of dysphagia over time and an increase in the number of the types of codes used.

The demographic information over time for the cohorts is recorded in Table 1, with stratified breakdowns for people without a dysphagia diagnosis recorded in Table 2 and those with a dysphagia diagnosis recorded in Table 3. The change in demographic information over time is also displayed in Fig. 1. Tables 2, 3 and Fig. 1 suggest differences in age, frailty (eFI) and WIMD between those with and without a dysphagia diagnosis, but no significant difference in gender. Table S3 includes the p values for chi-squared tests for differences in proportions for age, frailty (eFI), gender and WIMD. With the exception of gender, all tests showed significant differences in distributions for individuals with a dysphagia diagnosis compared to those without in the majority of the years tested (2008, 2010–2012, 2014–2015, and 2017–2018).

The regression results are displayed in Table 4. Multivariate analyses indicated increased odds ratios [OR (95% confidence intervals)] for a dysphagia diagnosis for increased age [reference 65–74: aged 75–84 OR 1.09 (1.07, 1.12), 85 + OR 1.23 (1.20, 1.27)], frailty [reference fit: mild frailty 2.45 (2.38, 2.53), moderate frailty 4.64 (4.49, 4.79) and severe frailty 7.87 (7.55, 8.21)], and increased ORs for the most deprived areas [reference 5. Least deprived, 1. Most deprived: 1.10 (1.06, 1.14)]. In the univariate analysis, males had a decreased odds of dysphagia diagnosis [OR 0.97 (0.95, 0.99)] but increased odds in the multivariate analysis [OR 1.14 (1.11, 1.16)]. The null model has a small, but non-zero, variance at the cohort year level, which suggests that there is variance between the cohort years.

The independent multilevel logistic regression results for the dysphagia diagnoses identified in primary and secondary care are displayed in the supplementary material, Tables S4 and S5, respectively. Results were consistent with the combined analysis but showed an increased odds of a recording of a dysphagia diagnosis in secondary care for those in both the most and second most deprived areas compared to the least deprived with ORs of 1.22 (1.17, 1.28) and 1.08 (1.03, 1.13), respectively.

We included only individuals with a confirmed general practice registration on the cohort index date. We were unable to include 100% coverage of the Welsh population due to some general practices not contributing data to the SAIL databank. As with many data linkage studies, we did not have perfect matching between primary care and administrative records which meant some individuals were lost when including demographic information. In addition, we acknowledge the limitations of a diagnosis of dysphagia being recorded in electronic health records as it may not be a primary diagnosis, there may be errors in coding and in common with many studies, there is no clear definition of dysphagia.

The information collected and documented in medical records is limited by the need for specific codes which may not be sensitive enough to accurately represent medical problems. Health care professionals do not routinely ask or screen for swallowing [55] even if there are well-documented neurological disease such as stroke, dementia or Parkinson’s disease.

There is no accepted, unifying definition of dysphagia, and this study has used a series of read codes and ICD-10 codes to capture swallowing problems/dysphagia. Some of these codes may not directly relate to oropharyngeal dysphagia.

We performed a large longitudinal data linkage study of over 400,000 individuals per year. This adds significant and novel evidence for the levels of recording of dysphagia in older adults. We were able to link primary and secondary care data together with administrative data to gain insights to the associations between dysphagia and the included demographics.