- Email: [email protected]
Data matters
OCCASIONAL REVIEW
Data matters
The advent of electronic patient records provides an opportunity to record data in a structured way that can give useful clinical information back to the clinician. This article aims to provide an understanding of how data recording takes place at the doctor epatient interface, and how data can be used to inform quality improvement. The difference between disease classification and terminology will be explained, as will the move towards “collecting once and using many times”. The advantages for clinicians and patients of improved data collection through the use of a single clinical terminology, The Systematised Nomenclature of Medicine Clinical Terms (SNOMED CT) will be demonstrated.
Stephen Andrew Spencer Ronny Cheung
Abstract Data underlie all service development and quality improvement initiatives in healthcare. This article describes the myriad potential benefits for improving current disjointed data systems through the creation of truly comprehensive electronic records. These will allow both real-time recording of the patientedoctor interaction and the ongoing process of care, as well as compatibility across healthcare settings and international boundaries. An important step is to implement a common clinical terminology to allow capture and sharing of clinical information with sufficient precision. SNOMED CT is the only system currently available with the potential to be healthcare data’s universal language. As with all data systems, the technology infrastructure is simply a means to improve health information delivery for clinicians, commissioners and policymakers. Its ultimate value to patients will depend on how healthcare professionals collect and use the information. It is a professional responsibility to ensure accurate and timely data collection through a thorough knowledge of health informatics as it applies at the doctorepatient interface.
Administrative healthcare data in England Most healthcare systems lack the NHS’s unified organisational responsibility for data, the infrastructure for collection, or the analytical capability for interpretation and use. As a result, healthcare data in England provides a rich source of information which is the envy of healthcare systems across the world. NHS data consists of millions of healthcare episodes over decades, from purely administrative data to richly detailed clinical information on diagnosis, procedures and treatments in many different healthcare settings. Despite this, healthcare information is currently created by an administrative process away from the point of patient care and clinicians have often felt detached from the process and frustrated when they wish to get involved. The current system is based on extraction of data from paper records. Inpatient diagnoses and treatments are extracted from unstructured notes by clinical coders and classified using the International Classification of Disease v 10 (ICD-10) and the Office of Population Census and Surveys Classification of Interventions and Procedures v 4 (OPCS-4) respectively. This data, along with a great deal of administrative detail relating to the admission, is submitted according to the requirements defined by the Clinical Data Set (CDS6.2) to an online database entitled Secondary User Services (SUS). The composite data is processed by the Health and Social Care Information Service (HSCIC) and subsequently published in various formats, most particularly as the Hospital Episode Statistics (HES). Currently, outpatient activity is not routinely coded for clinical information, including presenting symptoms, diagnosis or treatment. This same data underlies many different measures of healthcare performance. This includes Dr Foster’s “My Hospital Guide”, which published the Hospital Standardised Mortality Ratio (HSMR) that first drew attention to problems in MidStaffordshire NHS Trust, and ultimately to the publication of the Francis report. Hospital statistics are also used to calculate Standardised Hospital Mortality Index (SHMI) and to publish the recent surgical outcome data on NHS Choices showing mortality rates by surgeon. Secondary care coding was not designed to capture patientspecific detail at the doctorepatient interface e rather, it was based on diseases (and specific diagnostic or therapeutic interventions), and designed to give population-level epidemiological data for healthcare researchers and policymakers to plan services. By contrast, primary care data use different sets of clinical coding terminology to facilitate data collection at the doctor epatient interface - most commonly Read Codes Version 2,
Keywords electronic health record; electronic patient record; health informatics; hospital episode statistics; quality improvement; SNOMED CT
Introduction It is a truly shocking statistic that children in the UK are more likely to die in childhood than in almost any other developed country. This is known, only because of the diligence with which deaths are recorded in the developed world. But improvement requires more detailed information about the causes and circumstances surrounding these deaths. Is there a failure to prevent the causes of death, to provide adequate treatment, or both? Quality of healthcare cannot be measured solely by death rates. There have been various attempts across Europe and in the UK to develop outcome frameworks which rely heavily on the collection of healthcare statistics. The present system of collecting clinical data is designed on the basis that the information is spread across a paper record.
Stephen Andrew Spencer B.Med.Sci BM BS MRCP (UK) DM (Nottm) FRCPCH is a Retired Consultant Paediatrician and Neonatologist, University Hospital of North Staffordshire and National Clinical Lead for Hopsital Specialties, Health & Social Care Information Centre, Leeds, UK. Conflict of interest: Dr Spencer is employed for 1 day per week as National Clinical Lead for Hospital Specialities at the UK Terminology Centre of the Health and Social Care Information Centre. Ronny Cheung BM BCh MA MRCPCH PgDipMedEd is Consultant General Paediatrician, Evelina London Children’s Hosital, Guy’s and St Thomas’ NHS Foundation Trust, London and Clinical Advisor, Child and Maternal Health Intelligence Network, Public Health England, UK. Conflict of interest: none declared.
PAEDIATRICS AND CHILD HEALTH --:-
1
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
not helpful to their purpose. An analogy can be made with other classification systems, such as zoological taxonomy. For example, it would be perfectly logical to group together Lions, Tigers and Domestic Cats because they all belong to the same family. However in an international comparison of the outcome from injuries caused by cats, countries dealing mainly with injuries from big cats would inevitably have worse outcomes that could be attributed unfairly to the quality of care. Unless the system is detailed enough to allow analysis of the data using more meaningful criteria (in this case, species of Cat) the data cannot be retrospectively tooled for other purposes. Creating meaningful clinical datasets for improvement and research requires the recording of multiple levels of information, representing the multiple taxonomies for each individual diagnosis or symptom. SNOMED CT is a system which makes this possible.
developed by James Read, a General Practitioner from Loughborough. This is the basis of primary care monitoring, including the Quality Outcomes Framework (QOF) by which a significant proportion of GP remuneration is managed. Although these data are collected routinely by all practices, several non-compatible electronic systems are in use. Therefore comprehensive data for primary care services in England are laborious to analyse retrospectively, except for relatively small, curated research databases such as The Health Improvement Network (THIN) or Clinical Practice Research Datalink (CPRD). Lack of interoperability between primary care and hospital data means patient-level data linkage is not routinely possible.
Healthcare data for quality improvement It was Don Berwick, a paediatrician working in North America, who first drew attention to variations in patient safety across healthcare systems. Collection and use of good quality data is the key to making hospitals and clinical care better and safer.
SNOMED CT Quality improvement and audit require very specific information about diagnosis and treatment. Using a disease classification such as ICD-10 to group similar conditions makes the data more manageable for large scale analysis, but a great deal of important detail is lost. Conversely most clinicians are very specific when writing in free text. In fact the medical profession has a long history of eloquent descriptions of disease which helped to take forward medicine before the pathophysiology of many conditions was understood. The description by Sir Frederic Still of Pyloric Stenosis is a prime example and review of the notes in such a case would leave no doubt about the surety of the diagnosis.
“..it is clear that only what can be measured can be improved” - Lord Darzi Given the clear importance of data, the involvement of clinicians in collecting and reviewing data about the care of patients would be expected. When the current system was first launched in 1989 following the report by Edith Korner in 1982 the president of the Royal College of Physicians said: “The recommendation that diagnostic data should be collected on all patients covered by the system is to be welcomed; its omission would make the scheme even more obviously a management exercise, thereby lessening its appeal to the active clinician. Both for the sake of analysing the use of his own unit and for the sake of colleagues in epidemiology, however, he should accept the responsibility of making the diagnostic coding as accurate as possible” - Douglas Black
“Since the vomiting began the bowels have been costive, perhaps only opened with enemata. And now the infant is wasting to a marked degree and perhaps it is this wasting rather than any alarm at the vomiting which leads the parents to seek medical advice. Such is the history which leads one to examine specially for the two characteristic signs e visible and very marked peristalsis of the stomach and a palpable thickening of the pylorus e upon which the diagnosis rests.” The same issue exists for procedures where OPCS-4 is used to classify procedures, which are further grouped into Human Resource Groups (HRGs) for the purpose of commissioning and payment. Clinicians in highly technical specialties may get heavily involved in the coding of their procedures because even small changes in the classification can result in a large increase in departmental income. However, frustration exists as the classification does not support the detail that many clinicians require to record the complexity of their work, especially where these data are used to report outcomes linked to individual teams or clinicians. In preparation for electronic patient records, the NHS has been developing a structured terminology that is suitable for clinicians to record the details of every aspect of their clinical work. SNOMED CT was born in 1999 from an amalgamation of an American Pathology set of terms (SNOMED-RT) and Read Clinical Terms v3. SNOMED CT has been mandated for use for a number of years and is part of all secondary care electronic patient records (EPR) deployed through the National Programme for IT. It has been in use in a number of hospitals for well over 4 years. Since the
Sadly clinical involvement with national data collection has never been strong. A national survey of consultants found that only 22% had regular involvement with clinical coding and 36% thought it was important but did not involve them. The situation is entirely different where clinicians have led the development of bespoke data collection systems, such as the Badger System in neonatology, or have worked with national data collection, as in diabetes care. These enterprises highlight the potential for improvement when clinical interest is evoked. However successful, these systems are expensive and will never capture more than a tiny portion of NHS encounters. One of the reasons clinicians have been reluctant to involve themselves in the current system of national data collection is that they often find it difficult to access the data on their own patients. “.there are no immediate benefits for those who use them [clinical data].” - Audit Commission Report When clinicians have accessed hospital data they have often been disappointed because the diagnostic and procedure data are aggregated for coding purposes into groups of similar (or not so similar) conditions. This is often interpreted as inaccurate data, whereas it is actually a result of aggregating data in a way that is
PAEDIATRICS AND CHILD HEALTH --:-
2
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
Department of Health announced an aspiration for all Trusts to have an EPR by 2018 the uptake of the terminology has significantly increased, and the UK Terminology Centre (UKTC) is supporting SNOMED CT in both hospital and primary care records. So what exactly is SNOMED CT and how is it different to a disease classification? A disease classification or clinical coding system is designed to group similar conditions in a useful and statistically valid way. It requires arcane rules, a dedicated team of clinical coders who have a unique understanding of how the rules apply, and medical records to be kept in good order. Once coded, detail in the source data is lost, so if the grouping is not helpful for a given application such as medical audit, the source data has to be accessed again. SNOMED CT, which is a structured terminology, is designed to be a flexible vocabulary that allows for the precise description of any medical condition or procedure, as well as symptoms, medicines, clinical observations etc. It is updated regularly and is suitable for use by clinicians at the doctorepatient interface and for all forms of medical communication. Maps exist which allow appropriate SNOMED CT terms to be translated to ICD-10 or OPCS-4 on a many-to-one basis. Retaining the granularity of the original data means that the data can be analysed in different ways for different requirements. Providing a comprehensive understanding of the structure of SNOMED CT cannot be achieved in a few words and is therefore beyond the scope of this article. The interested reader is referred to the readily available online training materials (See Further reading). From a clinician’s perspective, the essential point is that each SNOMED concept, which may be a symptom, diagnosis, procedure or one of many other health related concepts, represents a unique clinical thought. As such it has a unique concept identification number. That concept can be described by one or more descriptions: for example carpal tunnel syndrome is also referred to as median nerve entrapment. Each concept has a fully specified name (a unique description that includes the type of term e.g. procedure), a preferred term (most common description) and also any number of synonyms. Each concept is placed in a hierarchy, so that there are “parent” terms which are more general and lie higher in the taxonomy, and “children” terms which are more specific and lie lower in the taxonomy. For instance, bacterial pneumonia has many synonyms (e.g. bacterial lower respiratory infection; lobar pneumonia), has a parent term (e.g. infective pneumonia) and many children terms (e.g. staphylococcal pneumonia). A concept may appear in more than one relationship: in this example, staphylococcal pneumonia can be tracked back to both a disease of the respiratory system and an infectious disease. This makes subsequent data analysis much easier as it is not necessary to search for every single concept within the area of interest. SNOMED CT contains a very large number of terms as it caters for every specialty, every health environment and every type of health professional. All 400,000 terms can be searched easily and rapidly, assuming that the ideal term is available and can be distinguished from other similar terms. Medical language is full of eponyms, synonyms, outdated concepts, overlapping terms and poorly defined entities. One paediatrician might prefer vesico-ureteric reflux with renal scarring, another chronic pyelonephritis, but do these two terms describe the same condition?
PAEDIATRICS AND CHILD HEALTH --:-
Pilot projects within the disciplines of rheumatology and paediatric disability have retired, corrected or placed in a different relationship existing terms and requested new terms to embrace modern understanding of disease. In SNOMED CT outdated concepts can be retired from ongoing use, while being retained in historical records for retrospective analyses. It has been argued that the development of terminology is unnecessary as powerful software is able to abstract clinical terms in a context sensitive way from free text. As technology develops, information retrieval systems will undoubtedly have a role to play. However, the problem is not in the systems, but in the use and consistency of language. Unless the profession and individual specialties come to an agreement on consistent use of medical language, with agreed definitions, data will always be of uncertain quality. SNOMED CT is supported by expert terminologists who can help clinicians to get to grips with the structure and concepts. Owing to the strength and variety of opinion regarding clinical terms, it is essential that expert clinical groups advising the development of content in SNOMED CT have a national mandate from their professional body. There is also a continuing requirement to maintain terminology as new disease concepts are developed and old ones retired. How can SNOMED CT improve care? Imagine you’re a general paediatrician, with a particular interest in a subspecialty. You work closely with your regional network which provides excellent clinical guidelines, care pathways and shared care arrangements. But attempts to develop credible quality indicators for monitoring purposes fail owing to inability to collect the detailed information required. How can better healthcare information systems, using SNOMED CT as a common language, improve the care you provide to your patients, now and in future? Comprehensive, shared EPR at the point of care Many of us have shared the frustrations of patients whose results from clinical reviews or investigations in another primary or secondary care environment are unavailable owing to the preponderance of paper-based systems. Important communication is currently dependent on typed letters and faxes. A common clinical terminology such as SNOMED CT would facilitate the creation of a single patient record across all settings. A general paediatrician caring for a child with medically complex needs would be able to quickly access relevant information from the patient’s comprehensive shared record, which may include inpatient and outpatient clinical review and results from other specialties (e.g. orthopaedics, surgery and radiology) or other healthcare providers (e.g. mental health or primary care). Clinical details of the current consultation can be added contemporaneously, which will be available to any other providers to view e potentially even if the patient moved abroad. The seamless inter-professional communication provided by such a summary care record would be impossible without a single shared terminology. This advance has implications for efficiency and patient safety, by bringing to the forefront pre-existing clinical data relevant to the current episode of care. In anaesthetics, the consultation record could be pre-populated with relevant patient
3
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
records across healthcare settings may enable this to be implemented prospectively to support and alert clinicians in management decisions. For instance, a patient may be recorded as suffering from recurrent mouth ulcers in a GP consultation, a raised ESR on an unrelated attendance to emergency department, and joint pains in a previous rheumatology clinic. A gastroenterologist may be able to set up trigger tools which would alert them, in real time, to all such patients with that particular constellation of results, and advise the GP to refer that patient for an assessment. The degree of detail available in SNOMED CT gives clinicians the ability to interrogate their patient population in a very specific way. A paediatric rheumatologist could build a case for specialist physiotherapy for shoulder pain using data from SNOMED CT, which would simply not be possible using current data systems or codes. Indeed, the detail is such that if a hypothetical new drug became available to treat all patients with right thumb pain, SNOMED CT would allow identification and data extraction for all of those patients, across all settings, so that they can be contacted and offered the treatment. Currently, such changes in treatment would have to rely on individual clinicians’ own data systems and their own system of dissemination, resulting in variation and delay in uptake.
risks (e.g. previous drug allergy, recent haemoglobin level) or allow pre-set health conditions to trigger an alert (e.g. previous difficult intubation). In community paediatrics, a cross-provider electronic patient record could bring forward relevant medical and social history for each new patient. This might include perinatal history (from a neonatal unit admission), immunisation history (from primary care), previous orthopaedic assessments or blood tests (hospital data) and at risk status (social care). This obviates the need for trawling through previous clinical correspondence (which may be misfiled or misplaced), and for repeating investigations or reviews. Population health care Health services are embracing the concept of population health. Traditionally, a paediatrician’s service relied upon the steady stream of referrals, mainly from primary or community services. Few looked beyond the walls of the hospital to assess whether the population’s needs were reflected in the demand for their service. Attempts to do so were frustrated by incomplete data and lack of patient information linked between primary, emergency department and secondary care. Creation of shared records between healthcare providers, using a shared terminology, facilitates population healthcare for hospital specialists. Information on children within local primary care services (or emergency department attendances) would be accessible, so patients who might benefit from secondary care management can be “pulled” into an appropriate clinic. A dietetic service could highlight all children in their catchment area on special feeds or who gain or lose weight inappropriately. A community developmental team could create triggers to alert them to children who have a diagnosis recorded in hospital which requires a time critical referral to support services. For commissioners of healthcare, a unified clinical record provides greater level of detail for service planning. It opens the possibility of linking two sets of data points together to assess quality and appropriateness of care. Commissioners could determine whether evidence-based guidelines are robustly followed for surgical or diagnostic procedure (e.g. recurrent tonsillitis or documented obstructive sleep apnoea for adenotonsillectomy) and whether they are linked to improved outcomes (e.g. reduced primary care attendances with tonsillitis). With current information systems, these scenarios would require data sharing agreements between several providers, most of whom use incompatible systems and codes. Some clinicians create their own databases to store this information, but these will be categorised idiosyncratically. A manual search would be required under different terminology for each dataset, with no provision for repeating such a resource-heavy exercise at regular intervals.
Quality indicators, national audit and variation in outcomes Quality Indicators are designed to allow outcomes to be carefully monitored for the purposes of patient safety and improvement. Even the simplest ideas, such as admission rates for children with epilepsy following discharge from paediatric clinic, require detailed diagnostic information if comparisons between hospitals or areas are to be meaningful. National audit requires a separate data collection in order to collect the required details, which is expensive and time-consuming. National analyses of clinical outcomes using administrative healthcare data, as in the NHS Atlas of Variation in Healthcare for Children and Young People, have highlighted geographical variations in quality, access and outcomes in England. Exploring the causes of this variation has been limited by the lack of detailed clinical data available. Introduction of SNOMED CT offers a potential solution. Research implications The use of healthcare data for research is an increasingly fruitful area. Large healthcare databases, comprising many millions of data points, have the potential to transform healthcare. They can complement much more expensive and labour-intensive clinical trials. They can also provide evidence for the effectiveness of care in situations where clinical trials are neither feasible nor possible e for instance, the impact of how health policies are implemented, or how health systems deliver care. Unfortunately these data are currently limited by the lack of microscopic clinical detail which could be provided by the 400,000 terms in SNOMED CT. The adaptive nature of SNOMED CT means that terms which have been changed can still be searched retrospectively. Consequently, longitudinal data can be more effectively studied without changing the source data, or compromising on the study population. International comparison is becoming increasingly important. Lack of a universal clinical coding language makes linkage across
Clinical decision support and patient trigger tools Detailed electronic patient records using standardised terminology raise the possibility of sophisticated clinical decision support for point of care use. The potential for electronic records as trigger tools for clinicians is already being explored. Existing algorithms can retrospectively identify potentially delayed diagnosis of cancer in patients presenting with a certain constellation of symptoms, clinical findings and investigation results. Shared
PAEDIATRICS AND CHILD HEALTH --:-
4
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
9 Spencer A, Modi N. National neonatal data to support specialist care and improve infant outcomes. Arch Dis Child Fetal Neonatal Ed 2013; 98: F175e80. 10 Spencer SA, Davies MP. Hospital episode statistics: improving the quality and value of hospital data: a national internet e-survey of hospital consultants. BMJ Open 2012; 2.
healthcare databases extremely challenging. SNOMED CT is an international terminology which is already available in English (both US and UK), Spanish, French, Swedish and Danish, and work is underway to extend to other languages. Robust studies of international healthcare variation would become one step closer to reality using SNOMED CT as a shared terminology.
Sources of funding
Key learning points
There were no sponsors or sources of funding for this article.A C
FURTHER READING 1 Berwick DM. Errors today and errors tomorrow. N Engl J Med 2003; 348: 2570e2. 2 Children and Young Peoples Health Outcomes Forum. Report of the Children and Young Peoples Health Outcomes Forum 2012. http:// www.dh.gov.uk/health/files/2012/07/CYP-report.pdf. 3 Gaywood I, Pande I. Preparing for electronic health records e Standardizing Terminology. Rheumatology 2014; http://dx.doi.org/10. 1093/rheumatology/keu383. 4 HSCIC; Summary Hospital e Level mortality indicator. http://www. hscic.gov.uk/SHMI. 5 IHTSDO. SNOMED. CT Starter guide. 2014. Available from: URL: http://ihtsdo.org/fileadmin/user_upload/doc/download/doc_ StarterGuide_Current-en-US_INT_20140731.pdf?ok. 6 IHTSDO. SNOMED CT E-Learning Centre. http://ihtsdo.org/fileadmin/ user_upload/doc/elearning.html. 7 National Child & Maternal Health Intelligence Network. NHS atlas of variation in healthcare for children and young people. http://www. chimat.org.uk/variation. 8 Spencer A. The white papers, quality indicators and clinical responsibility. Clin Med 2012; 12: 1e19.
PAEDIATRICS AND CHILD HEALTH --:-
C
C
C
Clinicians need to understand the changes to data collection that will occur as a result of the implementation of electronic patient records. Clinicians’ responsibility to collect and use data to improve quality of services and patient safety requires an understanding of terminology and clinical coding systems. SNOMED CT is a universal clinical coding language which has sufficient detail, interoperability and flexibility to allow electronic records to fulfill their potential for delivering better care, and for healthcare research. Professional input into the development of consistent terminology in SNOMED CT is essential.
Acknowledgement We gratefully acknowledge the support of Denise Downs, Implementation and Education Lead at the UK Terminology Centre, who read through various drafts of the document to ensure technical accuracy. She also made helpful suggestions to improve the content.
5
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
Data matters
The advent of electronic patient records provides an opportunity to record data in a structured way that can give useful clinical information back to the clinician. This article aims to provide an understanding of how data recording takes place at the doctor epatient interface, and how data can be used to inform quality improvement. The difference between disease classification and terminology will be explained, as will the move towards “collecting once and using many times”. The advantages for clinicians and patients of improved data collection through the use of a single clinical terminology, The Systematised Nomenclature of Medicine Clinical Terms (SNOMED CT) will be demonstrated.
Stephen Andrew Spencer Ronny Cheung
Abstract Data underlie all service development and quality improvement initiatives in healthcare. This article describes the myriad potential benefits for improving current disjointed data systems through the creation of truly comprehensive electronic records. These will allow both real-time recording of the patientedoctor interaction and the ongoing process of care, as well as compatibility across healthcare settings and international boundaries. An important step is to implement a common clinical terminology to allow capture and sharing of clinical information with sufficient precision. SNOMED CT is the only system currently available with the potential to be healthcare data’s universal language. As with all data systems, the technology infrastructure is simply a means to improve health information delivery for clinicians, commissioners and policymakers. Its ultimate value to patients will depend on how healthcare professionals collect and use the information. It is a professional responsibility to ensure accurate and timely data collection through a thorough knowledge of health informatics as it applies at the doctorepatient interface.
Administrative healthcare data in England Most healthcare systems lack the NHS’s unified organisational responsibility for data, the infrastructure for collection, or the analytical capability for interpretation and use. As a result, healthcare data in England provides a rich source of information which is the envy of healthcare systems across the world. NHS data consists of millions of healthcare episodes over decades, from purely administrative data to richly detailed clinical information on diagnosis, procedures and treatments in many different healthcare settings. Despite this, healthcare information is currently created by an administrative process away from the point of patient care and clinicians have often felt detached from the process and frustrated when they wish to get involved. The current system is based on extraction of data from paper records. Inpatient diagnoses and treatments are extracted from unstructured notes by clinical coders and classified using the International Classification of Disease v 10 (ICD-10) and the Office of Population Census and Surveys Classification of Interventions and Procedures v 4 (OPCS-4) respectively. This data, along with a great deal of administrative detail relating to the admission, is submitted according to the requirements defined by the Clinical Data Set (CDS6.2) to an online database entitled Secondary User Services (SUS). The composite data is processed by the Health and Social Care Information Service (HSCIC) and subsequently published in various formats, most particularly as the Hospital Episode Statistics (HES). Currently, outpatient activity is not routinely coded for clinical information, including presenting symptoms, diagnosis or treatment. This same data underlies many different measures of healthcare performance. This includes Dr Foster’s “My Hospital Guide”, which published the Hospital Standardised Mortality Ratio (HSMR) that first drew attention to problems in MidStaffordshire NHS Trust, and ultimately to the publication of the Francis report. Hospital statistics are also used to calculate Standardised Hospital Mortality Index (SHMI) and to publish the recent surgical outcome data on NHS Choices showing mortality rates by surgeon. Secondary care coding was not designed to capture patientspecific detail at the doctorepatient interface e rather, it was based on diseases (and specific diagnostic or therapeutic interventions), and designed to give population-level epidemiological data for healthcare researchers and policymakers to plan services. By contrast, primary care data use different sets of clinical coding terminology to facilitate data collection at the doctor epatient interface - most commonly Read Codes Version 2,
Keywords electronic health record; electronic patient record; health informatics; hospital episode statistics; quality improvement; SNOMED CT
Introduction It is a truly shocking statistic that children in the UK are more likely to die in childhood than in almost any other developed country. This is known, only because of the diligence with which deaths are recorded in the developed world. But improvement requires more detailed information about the causes and circumstances surrounding these deaths. Is there a failure to prevent the causes of death, to provide adequate treatment, or both? Quality of healthcare cannot be measured solely by death rates. There have been various attempts across Europe and in the UK to develop outcome frameworks which rely heavily on the collection of healthcare statistics. The present system of collecting clinical data is designed on the basis that the information is spread across a paper record.
Stephen Andrew Spencer B.Med.Sci BM BS MRCP (UK) DM (Nottm) FRCPCH is a Retired Consultant Paediatrician and Neonatologist, University Hospital of North Staffordshire and National Clinical Lead for Hopsital Specialties, Health & Social Care Information Centre, Leeds, UK. Conflict of interest: Dr Spencer is employed for 1 day per week as National Clinical Lead for Hospital Specialities at the UK Terminology Centre of the Health and Social Care Information Centre. Ronny Cheung BM BCh MA MRCPCH PgDipMedEd is Consultant General Paediatrician, Evelina London Children’s Hosital, Guy’s and St Thomas’ NHS Foundation Trust, London and Clinical Advisor, Child and Maternal Health Intelligence Network, Public Health England, UK. Conflict of interest: none declared.
PAEDIATRICS AND CHILD HEALTH --:-
1
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
not helpful to their purpose. An analogy can be made with other classification systems, such as zoological taxonomy. For example, it would be perfectly logical to group together Lions, Tigers and Domestic Cats because they all belong to the same family. However in an international comparison of the outcome from injuries caused by cats, countries dealing mainly with injuries from big cats would inevitably have worse outcomes that could be attributed unfairly to the quality of care. Unless the system is detailed enough to allow analysis of the data using more meaningful criteria (in this case, species of Cat) the data cannot be retrospectively tooled for other purposes. Creating meaningful clinical datasets for improvement and research requires the recording of multiple levels of information, representing the multiple taxonomies for each individual diagnosis or symptom. SNOMED CT is a system which makes this possible.
developed by James Read, a General Practitioner from Loughborough. This is the basis of primary care monitoring, including the Quality Outcomes Framework (QOF) by which a significant proportion of GP remuneration is managed. Although these data are collected routinely by all practices, several non-compatible electronic systems are in use. Therefore comprehensive data for primary care services in England are laborious to analyse retrospectively, except for relatively small, curated research databases such as The Health Improvement Network (THIN) or Clinical Practice Research Datalink (CPRD). Lack of interoperability between primary care and hospital data means patient-level data linkage is not routinely possible.
Healthcare data for quality improvement It was Don Berwick, a paediatrician working in North America, who first drew attention to variations in patient safety across healthcare systems. Collection and use of good quality data is the key to making hospitals and clinical care better and safer.
SNOMED CT Quality improvement and audit require very specific information about diagnosis and treatment. Using a disease classification such as ICD-10 to group similar conditions makes the data more manageable for large scale analysis, but a great deal of important detail is lost. Conversely most clinicians are very specific when writing in free text. In fact the medical profession has a long history of eloquent descriptions of disease which helped to take forward medicine before the pathophysiology of many conditions was understood. The description by Sir Frederic Still of Pyloric Stenosis is a prime example and review of the notes in such a case would leave no doubt about the surety of the diagnosis.
“..it is clear that only what can be measured can be improved” - Lord Darzi Given the clear importance of data, the involvement of clinicians in collecting and reviewing data about the care of patients would be expected. When the current system was first launched in 1989 following the report by Edith Korner in 1982 the president of the Royal College of Physicians said: “The recommendation that diagnostic data should be collected on all patients covered by the system is to be welcomed; its omission would make the scheme even more obviously a management exercise, thereby lessening its appeal to the active clinician. Both for the sake of analysing the use of his own unit and for the sake of colleagues in epidemiology, however, he should accept the responsibility of making the diagnostic coding as accurate as possible” - Douglas Black
“Since the vomiting began the bowels have been costive, perhaps only opened with enemata. And now the infant is wasting to a marked degree and perhaps it is this wasting rather than any alarm at the vomiting which leads the parents to seek medical advice. Such is the history which leads one to examine specially for the two characteristic signs e visible and very marked peristalsis of the stomach and a palpable thickening of the pylorus e upon which the diagnosis rests.” The same issue exists for procedures where OPCS-4 is used to classify procedures, which are further grouped into Human Resource Groups (HRGs) for the purpose of commissioning and payment. Clinicians in highly technical specialties may get heavily involved in the coding of their procedures because even small changes in the classification can result in a large increase in departmental income. However, frustration exists as the classification does not support the detail that many clinicians require to record the complexity of their work, especially where these data are used to report outcomes linked to individual teams or clinicians. In preparation for electronic patient records, the NHS has been developing a structured terminology that is suitable for clinicians to record the details of every aspect of their clinical work. SNOMED CT was born in 1999 from an amalgamation of an American Pathology set of terms (SNOMED-RT) and Read Clinical Terms v3. SNOMED CT has been mandated for use for a number of years and is part of all secondary care electronic patient records (EPR) deployed through the National Programme for IT. It has been in use in a number of hospitals for well over 4 years. Since the
Sadly clinical involvement with national data collection has never been strong. A national survey of consultants found that only 22% had regular involvement with clinical coding and 36% thought it was important but did not involve them. The situation is entirely different where clinicians have led the development of bespoke data collection systems, such as the Badger System in neonatology, or have worked with national data collection, as in diabetes care. These enterprises highlight the potential for improvement when clinical interest is evoked. However successful, these systems are expensive and will never capture more than a tiny portion of NHS encounters. One of the reasons clinicians have been reluctant to involve themselves in the current system of national data collection is that they often find it difficult to access the data on their own patients. “.there are no immediate benefits for those who use them [clinical data].” - Audit Commission Report When clinicians have accessed hospital data they have often been disappointed because the diagnostic and procedure data are aggregated for coding purposes into groups of similar (or not so similar) conditions. This is often interpreted as inaccurate data, whereas it is actually a result of aggregating data in a way that is
PAEDIATRICS AND CHILD HEALTH --:-
2
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
Department of Health announced an aspiration for all Trusts to have an EPR by 2018 the uptake of the terminology has significantly increased, and the UK Terminology Centre (UKTC) is supporting SNOMED CT in both hospital and primary care records. So what exactly is SNOMED CT and how is it different to a disease classification? A disease classification or clinical coding system is designed to group similar conditions in a useful and statistically valid way. It requires arcane rules, a dedicated team of clinical coders who have a unique understanding of how the rules apply, and medical records to be kept in good order. Once coded, detail in the source data is lost, so if the grouping is not helpful for a given application such as medical audit, the source data has to be accessed again. SNOMED CT, which is a structured terminology, is designed to be a flexible vocabulary that allows for the precise description of any medical condition or procedure, as well as symptoms, medicines, clinical observations etc. It is updated regularly and is suitable for use by clinicians at the doctorepatient interface and for all forms of medical communication. Maps exist which allow appropriate SNOMED CT terms to be translated to ICD-10 or OPCS-4 on a many-to-one basis. Retaining the granularity of the original data means that the data can be analysed in different ways for different requirements. Providing a comprehensive understanding of the structure of SNOMED CT cannot be achieved in a few words and is therefore beyond the scope of this article. The interested reader is referred to the readily available online training materials (See Further reading). From a clinician’s perspective, the essential point is that each SNOMED concept, which may be a symptom, diagnosis, procedure or one of many other health related concepts, represents a unique clinical thought. As such it has a unique concept identification number. That concept can be described by one or more descriptions: for example carpal tunnel syndrome is also referred to as median nerve entrapment. Each concept has a fully specified name (a unique description that includes the type of term e.g. procedure), a preferred term (most common description) and also any number of synonyms. Each concept is placed in a hierarchy, so that there are “parent” terms which are more general and lie higher in the taxonomy, and “children” terms which are more specific and lie lower in the taxonomy. For instance, bacterial pneumonia has many synonyms (e.g. bacterial lower respiratory infection; lobar pneumonia), has a parent term (e.g. infective pneumonia) and many children terms (e.g. staphylococcal pneumonia). A concept may appear in more than one relationship: in this example, staphylococcal pneumonia can be tracked back to both a disease of the respiratory system and an infectious disease. This makes subsequent data analysis much easier as it is not necessary to search for every single concept within the area of interest. SNOMED CT contains a very large number of terms as it caters for every specialty, every health environment and every type of health professional. All 400,000 terms can be searched easily and rapidly, assuming that the ideal term is available and can be distinguished from other similar terms. Medical language is full of eponyms, synonyms, outdated concepts, overlapping terms and poorly defined entities. One paediatrician might prefer vesico-ureteric reflux with renal scarring, another chronic pyelonephritis, but do these two terms describe the same condition?
PAEDIATRICS AND CHILD HEALTH --:-
Pilot projects within the disciplines of rheumatology and paediatric disability have retired, corrected or placed in a different relationship existing terms and requested new terms to embrace modern understanding of disease. In SNOMED CT outdated concepts can be retired from ongoing use, while being retained in historical records for retrospective analyses. It has been argued that the development of terminology is unnecessary as powerful software is able to abstract clinical terms in a context sensitive way from free text. As technology develops, information retrieval systems will undoubtedly have a role to play. However, the problem is not in the systems, but in the use and consistency of language. Unless the profession and individual specialties come to an agreement on consistent use of medical language, with agreed definitions, data will always be of uncertain quality. SNOMED CT is supported by expert terminologists who can help clinicians to get to grips with the structure and concepts. Owing to the strength and variety of opinion regarding clinical terms, it is essential that expert clinical groups advising the development of content in SNOMED CT have a national mandate from their professional body. There is also a continuing requirement to maintain terminology as new disease concepts are developed and old ones retired. How can SNOMED CT improve care? Imagine you’re a general paediatrician, with a particular interest in a subspecialty. You work closely with your regional network which provides excellent clinical guidelines, care pathways and shared care arrangements. But attempts to develop credible quality indicators for monitoring purposes fail owing to inability to collect the detailed information required. How can better healthcare information systems, using SNOMED CT as a common language, improve the care you provide to your patients, now and in future? Comprehensive, shared EPR at the point of care Many of us have shared the frustrations of patients whose results from clinical reviews or investigations in another primary or secondary care environment are unavailable owing to the preponderance of paper-based systems. Important communication is currently dependent on typed letters and faxes. A common clinical terminology such as SNOMED CT would facilitate the creation of a single patient record across all settings. A general paediatrician caring for a child with medically complex needs would be able to quickly access relevant information from the patient’s comprehensive shared record, which may include inpatient and outpatient clinical review and results from other specialties (e.g. orthopaedics, surgery and radiology) or other healthcare providers (e.g. mental health or primary care). Clinical details of the current consultation can be added contemporaneously, which will be available to any other providers to view e potentially even if the patient moved abroad. The seamless inter-professional communication provided by such a summary care record would be impossible without a single shared terminology. This advance has implications for efficiency and patient safety, by bringing to the forefront pre-existing clinical data relevant to the current episode of care. In anaesthetics, the consultation record could be pre-populated with relevant patient
3
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
records across healthcare settings may enable this to be implemented prospectively to support and alert clinicians in management decisions. For instance, a patient may be recorded as suffering from recurrent mouth ulcers in a GP consultation, a raised ESR on an unrelated attendance to emergency department, and joint pains in a previous rheumatology clinic. A gastroenterologist may be able to set up trigger tools which would alert them, in real time, to all such patients with that particular constellation of results, and advise the GP to refer that patient for an assessment. The degree of detail available in SNOMED CT gives clinicians the ability to interrogate their patient population in a very specific way. A paediatric rheumatologist could build a case for specialist physiotherapy for shoulder pain using data from SNOMED CT, which would simply not be possible using current data systems or codes. Indeed, the detail is such that if a hypothetical new drug became available to treat all patients with right thumb pain, SNOMED CT would allow identification and data extraction for all of those patients, across all settings, so that they can be contacted and offered the treatment. Currently, such changes in treatment would have to rely on individual clinicians’ own data systems and their own system of dissemination, resulting in variation and delay in uptake.
risks (e.g. previous drug allergy, recent haemoglobin level) or allow pre-set health conditions to trigger an alert (e.g. previous difficult intubation). In community paediatrics, a cross-provider electronic patient record could bring forward relevant medical and social history for each new patient. This might include perinatal history (from a neonatal unit admission), immunisation history (from primary care), previous orthopaedic assessments or blood tests (hospital data) and at risk status (social care). This obviates the need for trawling through previous clinical correspondence (which may be misfiled or misplaced), and for repeating investigations or reviews. Population health care Health services are embracing the concept of population health. Traditionally, a paediatrician’s service relied upon the steady stream of referrals, mainly from primary or community services. Few looked beyond the walls of the hospital to assess whether the population’s needs were reflected in the demand for their service. Attempts to do so were frustrated by incomplete data and lack of patient information linked between primary, emergency department and secondary care. Creation of shared records between healthcare providers, using a shared terminology, facilitates population healthcare for hospital specialists. Information on children within local primary care services (or emergency department attendances) would be accessible, so patients who might benefit from secondary care management can be “pulled” into an appropriate clinic. A dietetic service could highlight all children in their catchment area on special feeds or who gain or lose weight inappropriately. A community developmental team could create triggers to alert them to children who have a diagnosis recorded in hospital which requires a time critical referral to support services. For commissioners of healthcare, a unified clinical record provides greater level of detail for service planning. It opens the possibility of linking two sets of data points together to assess quality and appropriateness of care. Commissioners could determine whether evidence-based guidelines are robustly followed for surgical or diagnostic procedure (e.g. recurrent tonsillitis or documented obstructive sleep apnoea for adenotonsillectomy) and whether they are linked to improved outcomes (e.g. reduced primary care attendances with tonsillitis). With current information systems, these scenarios would require data sharing agreements between several providers, most of whom use incompatible systems and codes. Some clinicians create their own databases to store this information, but these will be categorised idiosyncratically. A manual search would be required under different terminology for each dataset, with no provision for repeating such a resource-heavy exercise at regular intervals.
Quality indicators, national audit and variation in outcomes Quality Indicators are designed to allow outcomes to be carefully monitored for the purposes of patient safety and improvement. Even the simplest ideas, such as admission rates for children with epilepsy following discharge from paediatric clinic, require detailed diagnostic information if comparisons between hospitals or areas are to be meaningful. National audit requires a separate data collection in order to collect the required details, which is expensive and time-consuming. National analyses of clinical outcomes using administrative healthcare data, as in the NHS Atlas of Variation in Healthcare for Children and Young People, have highlighted geographical variations in quality, access and outcomes in England. Exploring the causes of this variation has been limited by the lack of detailed clinical data available. Introduction of SNOMED CT offers a potential solution. Research implications The use of healthcare data for research is an increasingly fruitful area. Large healthcare databases, comprising many millions of data points, have the potential to transform healthcare. They can complement much more expensive and labour-intensive clinical trials. They can also provide evidence for the effectiveness of care in situations where clinical trials are neither feasible nor possible e for instance, the impact of how health policies are implemented, or how health systems deliver care. Unfortunately these data are currently limited by the lack of microscopic clinical detail which could be provided by the 400,000 terms in SNOMED CT. The adaptive nature of SNOMED CT means that terms which have been changed can still be searched retrospectively. Consequently, longitudinal data can be more effectively studied without changing the source data, or compromising on the study population. International comparison is becoming increasingly important. Lack of a universal clinical coding language makes linkage across
Clinical decision support and patient trigger tools Detailed electronic patient records using standardised terminology raise the possibility of sophisticated clinical decision support for point of care use. The potential for electronic records as trigger tools for clinicians is already being explored. Existing algorithms can retrospectively identify potentially delayed diagnosis of cancer in patients presenting with a certain constellation of symptoms, clinical findings and investigation results. Shared
PAEDIATRICS AND CHILD HEALTH --:-
4
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010
OCCASIONAL REVIEW
9 Spencer A, Modi N. National neonatal data to support specialist care and improve infant outcomes. Arch Dis Child Fetal Neonatal Ed 2013; 98: F175e80. 10 Spencer SA, Davies MP. Hospital episode statistics: improving the quality and value of hospital data: a national internet e-survey of hospital consultants. BMJ Open 2012; 2.
healthcare databases extremely challenging. SNOMED CT is an international terminology which is already available in English (both US and UK), Spanish, French, Swedish and Danish, and work is underway to extend to other languages. Robust studies of international healthcare variation would become one step closer to reality using SNOMED CT as a shared terminology.
Sources of funding
Key learning points
There were no sponsors or sources of funding for this article.A C
FURTHER READING 1 Berwick DM. Errors today and errors tomorrow. N Engl J Med 2003; 348: 2570e2. 2 Children and Young Peoples Health Outcomes Forum. Report of the Children and Young Peoples Health Outcomes Forum 2012. http:// www.dh.gov.uk/health/files/2012/07/CYP-report.pdf. 3 Gaywood I, Pande I. Preparing for electronic health records e Standardizing Terminology. Rheumatology 2014; http://dx.doi.org/10. 1093/rheumatology/keu383. 4 HSCIC; Summary Hospital e Level mortality indicator. http://www. hscic.gov.uk/SHMI. 5 IHTSDO. SNOMED. CT Starter guide. 2014. Available from: URL: http://ihtsdo.org/fileadmin/user_upload/doc/download/doc_ StarterGuide_Current-en-US_INT_20140731.pdf?ok. 6 IHTSDO. SNOMED CT E-Learning Centre. http://ihtsdo.org/fileadmin/ user_upload/doc/elearning.html. 7 National Child & Maternal Health Intelligence Network. NHS atlas of variation in healthcare for children and young people. http://www. chimat.org.uk/variation. 8 Spencer A. The white papers, quality indicators and clinical responsibility. Clin Med 2012; 12: 1e19.
PAEDIATRICS AND CHILD HEALTH --:-
C
C
C
Clinicians need to understand the changes to data collection that will occur as a result of the implementation of electronic patient records. Clinicians’ responsibility to collect and use data to improve quality of services and patient safety requires an understanding of terminology and clinical coding systems. SNOMED CT is a universal clinical coding language which has sufficient detail, interoperability and flexibility to allow electronic records to fulfill their potential for delivering better care, and for healthcare research. Professional input into the development of consistent terminology in SNOMED CT is essential.
Acknowledgement We gratefully acknowledge the support of Denise Downs, Implementation and Education Lead at the UK Terminology Centre, who read through various drafts of the document to ensure technical accuracy. She also made helpful suggestions to improve the content.
5
Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Spencer SA, Cheung R, Data matters, Paediatrics and Child Health (2015), http://dx.doi.org/10.1016/ j.paed.2015.06.010