Patient outcomes following emergency admission to hospital for COVID-19 compared with influenza: retrospective cohort study.

BACKGROUND: We examine differences in posthospitalisation outcomes, and health system resource use, for patients hospitalised with COVID-19 during the UK's first pandemic wave in 2020, and influenza during 2018 and 2019. METHODS: This retrospective cohort study used routinely collected primary and secondary care data. Outcomes, measured for 90 days follow-up after discharge were length of stay in hospital, mortality, emergency readmission and primary care activity. RESULTS: The study included 5132 patients admitted to hospital as an emergency, with COVID-19 and influenza cohorts comprising 3799 and 1333 patients respectively. Patients in the COVID-19 cohort were more likely to stay in hospital longer than 10 days (OR 3.91, 95% CI 3.14 to 4.65); and more likely to die in hospital (OR 11.85, 95% CI 8.58 to 16.86) and within 90 days of discharge (OR 7.92, 95% CI 6.20 to 10.25). For those who survived, rates of emergency readmission within 90 days were comparable between COVID-19 and influenza cohorts (OR 1.07, 95% CI 0.89 to 1.29), while primary care activity was greater among the COVID-19 cohort (incidence rate ratio 1.30, 95% CI 1.23 to 1.37). CONCLUSIONS: Patients admitted for COVID-19 were more likely to die, more likely to stay in hospital for over 10 days and interact more with primary care after discharge, than patients admitted for influenza. However, readmission rates were similar for both groups. These findings, while situated in the context of the first wave of COVID-19, with the associated pressures on the health system, can inform health service planning for subsequent waves of COVID-19, and show that patients with COVID-19 interact more with healthcare services as well as having poorer outcomes than those with influenza.

Factors associated with accessing long-term adult social care in people aged 75 and over: a retrospective cohort study.

BACKGROUND: An ageing population and limited resources have put strain on state provision of adult social care (ASC) in England. With social care needs predicted to double over the next 20 years, there is a need for new approaches to inform service planning and development, including through predictive models of demand. OBJECTIVE: Describe risk factors for long-term ASC in two inner London boroughs and develop a risk prediction model for long-term ASC. METHODS: Pseudonymised person-level data from an integrated care dataset were analysed. We used multivariable logistic regression to model associations of demographic factors, and baseline aspects of health status and health service use, with accessing long-term ASC over 12 months. RESULTS: The cohort comprised 13,394 residents, aged ≥75 years with no prior history of ASC at baseline. Of these, 1.7% became ASC clients over 12 months. Residents were more likely to access ASC if they were older or living in areas with high socioeconomic deprivation. Those with preexisting mental health or neurological conditions, or more intense prior health service use during the baseline period, were also more likely to access ASC. A prognostic model derived from risk factors had limited predictive power. CONCLUSIONS: Our findings reinforce evidence on known risk factors for residents aged 75 or over, yet even with linked routinely collected health and social care data, it was not possible to make accurate predictions of long-term ASC use for individuals. We propose that a paradigm shift towards more relational, personalised approaches, is needed.

A modified Delphi study to identify the features of high quality measurement plans for healthcare improvement projects.

BACKGROUND: The design and execution of measurement in quality improvement (QI) initiatives is often poor. Better guidance on "what good looks like" might help to mitigate some of the problems. We report a consensus-building process that sought to identify which features are important to include in QI measurement plans. METHODS: We conducted a three-stage consensus-building approach: (1) identifying the list of features of measurement plans that were potential candidates for inclusion based on literature review and the study team's experience; (2) a two-round modified Delphi exercise with a panel of experts to establish consensus on the importance of these features; and (3) a small in-person consensus group meeting to finalise the list of features. RESULTS: A list of 104 candidate questions was generated. A panel of 19 experts in the Delphi reviewed these questions and produced consensus on retaining 46 questions in the first round and on a further 22 in the second round. Thematic analysis of open text responses from the panellists suggested a number of areas of debate that were explicitly considered by the consensus group. The exercise yielded 74 questions (71% of 104) on which there was consensus in five categories of measurement relating to: design, data collection and management, analysis, action, and embedding. CONCLUSIONS: This study offers a consensus-based view on the features of a good measurement plan for a QI project in healthcare. The results may be of use to QI teams, funders and evaluators, but are likely to require further development and testing to ensure feasibility and usefulness.

Understanding and extending the Starling principle.

The Starling Principle states that fluid movements between blood and tissues are determined by differences in hydrostatic and colloid osmotic (oncotic) pressures between plasma inside microvessels and fluid outside them. The Revised Starling Principle recognizes that, because microvessels are permeable to macromolecules, a balance of pressures cannot halt fluid exchange. In most tissues, steady oncotic pressure differences between plasma and interstitial fluid depend on low levels of steady filtration from plasma to tissues for which the Revised Principle provides the theory. Plasma volume is normally maintained by fluid losses from filtration being matched by fluid gains from lymph. Steady state fluid uptake into plasma only occurs in tissues such as intestinal mucosa and renal peri-tubular capillaries where a protein-free secretion of adjacent epithelia contributes significantly to interstitial fluid volume and keeps interstitial oncotic pressure low. Steady filtration rates in different tissues are disturbed locally by reflex changes in capillary pressure and perfusion. The rapid overall decline in capillary pressure after acute blood loss initiates rapid fluid uptake from tissue to plasma, that is, autotransfusion. Fluid uptake is transient, being rapid at first then attenuating but low levels may continue for more than an hour. The Revised Principle highlights the role of oncotic pressure of small volumes of interstitial fluid within a sub-compartment surrounding the microvessels rather than the tissue's mean interstitial fluid oncotic pressure. This maximizes oncotic pressure differences when capillary pressure are high and enhances initial absorption rates when pressures are low, accelerating short-term regulation of plasma volume.

FORWARD (Feeding via the Oral Route With Acknowledged Risk of Deterioration): evaluation of a novel tool to support patients eating and drinking at risk of aspiration.

BACKGROUND: care of patients with a permanently unsafe swallow who are inappropriate for tube feeding is challenging. Eating and drinking with acknowledged risk (EDAR) may be an appropriate strategy but without clear decision making and communication patients may spend unnecessarily long 'nil by mouth' (NBM), they or their family may experience significant anxieties and advance care plans may not be made. METHODS: the FORWARD (Feeding via the Oral Route With Acknowledged Risk of Deterioration) care bundle was sequentially co-designed and embedded across different in-patient clinical services using 'plan-do-study-act' methodology to systematise best practice. Care before and after FORWARD's implementation was evaluated using a time-series analysis of 305 'EDAR patients' (19 in 6 months pre-FORWARD; 42 in a 12-month 'pilot'; 244 patients in the subsequent 27 months). RESULTS: median (IQR) days patients were NBM without an alternative feeding route reduced significantly from 2 (1-4) pre-FORWARD to 0 (0-2) in the 'pilot' and 0 (0) post-'pilot' (P < 0.05). G-chart analysis demonstrated sustained performance across time and different clinical settings. Implementation of FORWARD significantly improved documentation of capacity assessment (42%→98%), discussions with next of kin (47%→98%) and onward communication of feeding plans (67%→81%). In wards where FORWARD was introduced, rate of aspiration pneumonia (a 'balancing measure' sensitive to harm associated with EDAR) increased at half the rate of dysphagia (0.8%/year versus 1.6%/year). CONCLUSION: the FORWARD care bundle supported complex decision-making around EDAR in patients with persistent dysphagia. The benefits of FORWARD were shown to be sustained over time and in a wide in-patient context.

Criteria for evaluating programme theory diagrams in quality improvement initiatives: a structured method for appraisal.

BACKGROUND: Despite criticisms that many quality improvement (QI) initiatives fail due to incomplete programme theory, there is no defined way to evaluate how programme theory has been articulated. The objective of this research was to develop, and assess the usability and reliability of scoring criteria to evaluate programme theory diagrams. METHODS: Criteria development was informed by published literature and QI experts. Inter-rater reliability was tested between two evaluators. About 63 programme theory diagrams (42 driver diagrams and 21 action-effect diagrams) were reviewed to establish whether the criteria could support comparative analysis of different approaches to constructing diagrams. RESULTS: Components of the scoring criteria include: assessment of overall aim, logical overview, clarity of components, cause-effect relationships, evidence and measurement. Independent reviewers had 78% inter-rater reliability. Scoring enabled direct comparison of different approaches to developing programme theory; action-effect diagrams were found to have had a statistically significant but moderate improvement in programme theory quality over driver diagrams; no significant differences were observed based on the setting in which driver diagrams were developed. CONCLUSIONS: The scoring criteria summarise the necessary components of programme theory that are thought to contribute to successful QI projects. The viability of the scoring criteria for practical application was demonstrated. Future uses include assessment of individual programme theory diagrams and comparison of different approaches (e.g. methodological, teaching or other QI support) to produce programme theory. The criteria can be used as a tool to guide the production of better programme theory diagrams, and also highlights where additional support for QI teams could be needed.

How to study improvement interventions: a brief overview of possible study types.

Improvement (defined broadly as purposive efforts to secure positive change) has become an increasingly important activity and field of inquiry within healthcare. This article offers an overview of possible methods for the study of improvement interventions. The choice of available designs is wide, but debates continue about how far improvement efforts can be simultaneously practical (aimed at producing change) and scientific (aimed at producing new knowledge), and whether the distinction between the practical and the scientific is a real and useful one. Quality improvement projects tend to be applied and, in some senses, self-evaluating. They are not necessarily directed at generating new knowledge, but reports of such projects if well conducted and cautious in their inferences may be of considerable value. They can be distinguished heuristically from research studies, which are motivated by and set out explicitly to test a hypothesis, or otherwise generate new knowledge, and from formal evaluations of improvement projects. We discuss variants of trial designs, quasi-experimental designs, systematic reviews, programme evaluations, process evaluations, qualitative studies, and economic evaluations. We note that designs that are better suited to the evaluation of clearly defined and static interventions may be adopted without giving sufficient attention to the challenges associated with the dynamic nature of improvement interventions and their interactions with contextual factors. Reconciling pragmatism and research rigour is highly desirable in the study of improvement. Trade-offs need to be made wisely, taking into account the objectives involved and inferences to be made.

Identifying reasons for delays in acute hospitals using the Day-of-Care Survey method.

This paper describes a new tool called 'Day-of-Care Survey', developed to assess inpatient delays in acute hospitals. Using literature review, iterative testing and feedback from professional groups, a national multidisciplinary team developed the survey criteria and methodology. Review teams working in pairs visited wards and used case records and bedside charts to assess the patient's status against severity of illness and service intensity criteria. Patients who did not meet the survey criteria for acute care were identified and delays were categorised. From March 2012 to December 2013, nine acute hospitals across Scotland, Australia and England were surveyed. A total of 3,846 adult general inpatient beds (excluding intensive care and maternity) were reviewed. There were 145 empty beds at the time of surveys across the nine sites, with 270 definite discharges planned on the day of the survey. The total number of patients not meeting criteria for acute care was 798/3,431 (23%, range 18-28%). Six factors accounted for 61% (490/798) of the reasons why patients not meeting acute care criteria remained in hospital. This survey gives important insights into the challenges of managing inpatient flow using system level information as a method to target interventions designed to address delay.

Model-driven approach to data collection and reporting for quality improvement.

Continuous data collection and analysis have been shown essential to achieving improvement in healthcare. However, the data required for local improvement initiatives are often not readily available from hospital Electronic Health Record (EHR) systems or not routinely collected. Furthermore, improvement teams are often restricted in time and funding thus requiring inexpensive and rapid tools to support their work. Hence, the informatics challenge in healthcare local improvement initiatives consists of providing a mechanism for rapid modelling of the local domain by non-informatics experts, including performance metric definitions, and grounded in established improvement techniques. We investigate the feasibility of a model-driven software approach to address this challenge, whereby an improvement model designed by a team is used to automatically generate required electronic data collection instruments and reporting tools. To that goal, we have designed a generic Improvement Data Model (IDM) to capture the data items and quality measures relevant to the project, and constructed Web Improvement Support in Healthcare (WISH), a prototype tool that takes user-generated IDM models and creates a data schema, data collection web interfaces, and a set of live reports, based on Statistical Process Control (SPC) for use by improvement teams. The software has been successfully used in over 50 improvement projects, with more than 700 users. We present in detail the experiences of one of those initiatives, Chronic Obstructive Pulmonary Disease project in Northwest London hospitals. The specific challenges of improvement in healthcare are analysed and the benefits and limitations of the approach are discussed.

Improving mental health outcomes: achieving equity through quality improvement.

OBJECTIVE: To investigate equity of patient outcomes in a psychological therapy service, following increased access achieved by a quality improvement (QI) initiative. DESIGN: Retrospective service evaluation of health outcomes; data analysed by ANOVA, chi-squared and Statistical Process Control. SETTING: A psychological therapy service in Westminster, London, UK. PARTICIPANTS: People living in the Borough of Westminster, London, attending the service (from either healthcare professional or self-referral) between February 2009 and May 2012. INTERVENTION: s) Social marketing interventions were used to increase referrals, including the promotion of the service through local media and through existing social networks. MAIN OUTCOME MEASURE: s) (i) Severity of depression on entry using Patient Health Questionnaire-9 (PHQ9). (ii) Changes to severity of depression following treatment (ΔPHQ9). (iii) Changes in attainment of a meaningful improvement in condition assessed by a key performance indicator. RESULTS: Patients from areas of high deprivation entered the service with more severe depression (M = 15.47, SD = 6.75), compared with patients from areas of low (M = 13.20, SD = 6.75) and medium (M = 14.44, SD = 6.64) deprivation. Patients in low, medium and high deprivation areas attained similar changes in depression score (ΔPHQ9: M = -6.60, SD = 6.41). Similar proportions of patients achieved the key performance indicator across initiative phase and deprivation categories. CONCLUSIONS: QI methods improved access to mental health services; this paper finds no evidence for differences in clinical outcomes in patients, regardless of level of deprivation, interpreted as no evidence of inequity in the service with respect to this outcome.

United Kingdom military aeromedical evacuation in the post-9/11 era.

BACKGROUND: Recent UK military operations in support of the fight against terrorism have resulted in UK military casualties. Movement of these casualties through the military medical chain requires a highly sophisticated aeromedical evacuation capability with worldwide reach. Recognition of the determinants of evacuation allows development to ensure optimal future configurations of military aeromedical evacuation services. METHODS: The database recording aeromedical evacuations undertaken by the Royal Air Force was searched to provide demographic and clinical data for evacuations between 1 April 2003 and 31 March 2010. Diagnoses leading to evacuation were categorized according to International Classification of Diseases codes. RESULTS: There were 21,477 medical evacuations undertaken. Analysis demonstrated 85.9% were for men and 86.5% were for military personnel, of whom 72.0% were in the army. The most common reasons for evacuation in military patients were musculoskeletal/connective tissue disorders (N = 9192; 50.0%), trauma (N = 1303; 7.1%), and mental health disorders (N = 1151; 6.3%). The most common reasons for evacuation in nonmilitary patients were musculoskeletal/connective tissue disorders (N = 734; 23.8%), genitourinary disorders (N = 325; 10.5%), and circulatory disorders (N = 255; 8.3%). Nontraumatic diagnoses were the determinants of evacuation in 92.9% of military and 95.1% of nonmilitary patients; 17.8% of trauma patients and 0.5% of nontrauma patients utilized high-dependency care. DISCUSSION: The UK aeromedical evacuation system must have the capacity to evacuate large numbers of patients with nontraumatic diagnoses, but also the flexibility to accommodate smaller, more variable numbers of higher dependency trauma patients. The military medical chain must continually review the differing requirements of civilian patients transferred within their aeromedical system.

Development of a Pollination Service Measurement (PSM) method using potted plant phytometry.

The value of pollination to human society is not limited to agricultural production, but also in the sustainability of ecosystems and the services that they provide. Seed set can be used as a comparative measure of pollination effectiveness, with minimum variability expected when other resources are not limiting. Six species of self-incompatible fall asters (Symphyotrichum) were used to evaluate pollination service at 12 sites across a spectrum of expected levels of pollination. Seed set per inflorescence was generally lower at sites with lower pollinator numbers and diversity, although as expected pollinator assemblage characteristics were highly variable within and between sites. However, rankings of sites showed consistency of response across phytometer species and between years; the summed ranks across multiple species appears to have as the greatest value in Pollination Service Measurement (PSM). Abundance, richness, and Shannon diversity of pollinator assemblages were highly autocorrelated and showed variable relationships with seed set depending on plant species and temporal scale of pollinator assemblage assessment. Use of seed set to directly measure pollination service at a site was consistent and cost effective when compared to less certain and more labour-intensive methods of pollinator collection and identification, and shows promise for implementation in pollination monitoring and bioassessment practices.

Effect of timeframes to define long term conditions and sociodemographic factors on prevalence of multimorbidity using disease code frequency in primary care electronic health records: retrospective study.

Objective: To determine the extent to which the choice of timeframe used to define a long term condition affects the prevalence of multimorbidity and whether this varies with sociodemographic factors. Design: Retrospective study of disease code frequency in primary care electronic health records. Data sources: Routinely collected, general practice, electronic health record data from the Clinical Practice Research Datalink Aurum were used. Main outcome measures: Adults (≥18 years) in England who were registered in the database on 1 January 2020 were included. Multimorbidity was defined as the presence of two or more conditions from a set of 212 long term conditions. Multimorbidity prevalence was compared using five definitions. Any disease code recorded in the electronic health records for 212 conditions was used as the reference definition. Additionally, alternative definitions for 41 conditions requiring multiple codes (where a single disease code could indicate an acute condition) or a single code for the remaining 171 conditions were as follows: two codes at least three months apart; two codes at least 12 months apart; three codes within any 12 month period; and any code in the past 12 months. Mixed effects regression was used to calculate the expected change in multimorbidity status and number of long term conditions according to each definition and associations with patient age, gender, ethnic group, and socioeconomic deprivation. Results: 9 718 573 people were included in the study, of whom 7 183 662 (73.9%) met the definition of multimorbidity where a single code was sufficient to define a long term condition. Variation was substantial in the prevalence according to timeframe used, ranging from 41.4% (n=4 023 023) for three codes in any 12 month period, to 55.2% (n=5 366 285) for two codes at least three months apart. Younger people (eg, 50-75% probability for 18-29 years v 1-10% for ≥80 years), people of some minority ethnic groups (eg, people in the Other ethnic group had higher probability than the South Asian ethnic group), and people living in areas of lower socioeconomic deprivation were more likely to be re-classified as not multimorbid when using definitions requiring multiple codes. Conclusions: Choice of timeframe to define long term conditions has a substantial effect on the prevalence of multimorbidity in this nationally representative sample. Different timeframes affect prevalence for some people more than others, highlighting the need to consider the impact of bias in the choice of method when defining multimorbidity.

Identifying multi-resolution clusters of diseases in ten million patients with multimorbidity in primary care in England.

BACKGROUND: Identifying clusters of diseases may aid understanding of shared aetiology, management of co-morbidities, and the discovery of new disease associations. Our study aims to identify disease clusters using a large set of long-term conditions and comparing methods that use the co-occurrence of diseases versus methods that use the sequence of disease development in a person over time. METHODS: We use electronic health records from over ten million people with multimorbidity registered to primary care in England. First, we extract data-driven representations of 212 diseases from patient records employing (i) co-occurrence-based methods and (ii) sequence-based natural language processing methods. Second, we apply the graph-based Markov Multiscale Community Detection (MMCD) to identify clusters based on disease similarity at multiple resolutions. We evaluate the representations and clusters using a clinically curated set of 253 known disease association pairs, and qualitatively assess the interpretability of the clusters. RESULTS: Both co-occurrence and sequence-based algorithms generate interpretable disease representations, with the best performance from the skip-gram algorithm. MMCD outperforms k-means and hierarchical clustering in explaining known disease associations. We find that diseases display an almost-hierarchical structure across resolutions from closely to more loosely similar co-occurrence patterns and identify interpretable clusters corresponding to both established and novel patterns. CONCLUSIONS: Our method provides a tool for clustering diseases at different levels of resolution from co-occurrence patterns in high-dimensional electronic health records, which could be used to facilitate discovery of associations between diseases in the future.

Having multiple long-term conditions is linked to worse health, poorer quality of life, and difficulties accessing healthcare. Identifying groups, or 'clusters' of diseases that are more likely to occur together in one person may help healthcare services to better meet the needs of those with multiple conditions. Our study aims to identify clusters of similar diseases, based not only on the diseases someone has now, but on the order in which they developed them. We compare a range of methods and find that our strategy performs best at explaining diseases that are already known to be linked, whilst also identifying new clusters of diseases. These methods could be used in future to better understand how diseases occur together, which could help the design of more efficient healthcare services.

Assigning disease clusters to people: A cohort study of the implications for understanding health outcomes in people with multiple long-term conditions.

Background: Identifying clusters of co-occurring diseases may help characterise distinct phenotypes of Multiple Long-Term Conditions (MLTC). Understanding the associations of disease clusters with health-related outcomes requires a strategy to assign clusters to people, but it is unclear how the performance of strategies compare. Aims: First, to compare the performance of methods of assigning disease clusters to people at explaining mortality, emergency department attendances and hospital admissions over one year. Second, to identify the extent of variation in the associations with each outcome between and within clusters. Methods: We conducted a cohort study of primary care electronic health records in England, including adults with MLTC. Seven strategies were tested to assign patients to fifteen disease clusters representing 212 LTCs, identified from our previous work. We tested the performance of each strategy at explaining associations with the three outcomes over 1 year using logistic regression and compared to a strategy using the individual LTCs. Results: 6,286,233 patients with MLTC were included. Of the seven strategies tested, a strategy assigning the count of conditions within each cluster performed best at explaining all three outcomes but was inferior to using information on the individual LTCs. There was a larger range of effect sizes for the individual LTCs within the same cluster than there was between the clusters. Conclusion: Strategies of assigning clusters of co-occurring diseases to people were less effective at explaining health-related outcomes than a person's individual diseases. Furthermore, clusters did not represent consistent relationships of the LTCs within them, which might limit their application in clinical research.

Treating insomnia in people who are incarcerated: a feasibility study of a multicomponent treatment pathway.

Around 60% of people who are incarcerated have insomnia; 6-10 times more prevalent than the general population. Yet, there is no standardized, evidence-based approach to insomnia treatment in prison. We assessed the feasibility of a treatment pathway for insomnia in a high-secure prison to inform a future randomized controlled trial (RCT) and initial efficacy data for sleep and mental health outcomes. We used a within-participants pre-post design. The stepped-care pathway included: self-management with peer support, environmental aids, and cognitive behavioral therapy for insomnia (CBTi). Assessment measures for insomnia, well-being, mood, anxiety, suicidality, overall health, sleepiness, fatigue, and cognitive functioning were administered at baseline and pathway exit. Feasibility criteria included eligibility to participate, CBTi uptake, and assessment completion. Forty-two adult males who are incarcerated were approached of which 95.2% were eligible. Of those deemed eligible, most participated (36/40, 90.0%). Most who completed baseline completed post-assessments (28/36, 77.8%) and of these, most showed improvements in their subjective sleep (27/28, 96.4%). Large reductions were found from pre- to posttreatment in insomnia severity (d = -1.81, 95% CI: 8.3 to 12.9) and 57.0% reported no clinically significant insomnia symptoms at post-assessment. There was no overall change in actigraphy-measured sleep. Large treatment benefits were found for depression, anxiety, well-being, and cognitive functioning, with a medium benefit on suicidal ideation. The treatment pathway for insomnia in prison was feasible and may be an effective treatment for insomnia in people who are incarcerated, with additional promising benefits for mental health. A pragmatic RCT across different prison populations is warranted. This paper is part of the Sleep and Circadian Health in the Justice System Collection.

Comparing natural language processing representations of coded disease sequences for prediction in electronic health records.

OBJECTIVE: Natural language processing (NLP) algorithms are increasingly being applied to obtain unsupervised representations of electronic health record (EHR) data, but their comparative performance at predicting clinical endpoints remains unclear. Our objective was to compare the performance of unsupervised representations of sequences of disease codes generated by bag-of-words versus sequence-based NLP algorithms at predicting clinically relevant outcomes. MATERIALS AND METHODS: This cohort study used primary care EHRs from 6 286 233 people with Multiple Long-Term Conditions in England. For each patient, an unsupervised vector representation of their time-ordered sequences of diseases was generated using 2 input strategies (212 disease categories versus 9462 diagnostic codes) and different NLP algorithms (Latent Dirichlet Allocation, doc2vec, and 2 transformer models designed for EHRs). We also developed a transformer architecture, named EHR-BERT, incorporating sociodemographic information. We compared the performance of each of these representations (without fine-tuning) as inputs into a logistic classifier to predict 1-year mortality, healthcare use, and new disease diagnosis. RESULTS: Patient representations generated by sequence-based algorithms performed consistently better than bag-of-words methods in predicting clinical endpoints, with the highest performance for EHR-BERT across all tasks, although the absolute improvement was small. Representations generated using disease categories perform similarly to those using diagnostic codes as inputs, suggesting models can equally manage smaller or larger vocabularies for prediction of these outcomes. DISCUSSION AND CONCLUSION: Patient representations produced by sequence-based NLP algorithms from sequences of disease codes demonstrate improved predictive content for patient outcomes compared with representations generated by co-occurrence-based algorithms. This suggests transformer models may be useful for generating multi-purpose representations, even without fine-tuning.