How predictive analytics can help find the rare disease patient
2 likes2,253 views
This document discusses how predictive analytics using real-world data can help identify undiagnosed rare disease patients. It describes two case studies: 1) A screening algorithm identified potentially undiagnosed patients for a rare multi-system disease with a high risk prevalence of 20.5% compared to 0.7% of the population. 2) An analysis of a rare cardiac disease identified health system barriers like variability between diagnostic centers that could cause under diagnosis. While initial results are promising, challenges remain around data privacy, sample size, and clinician adoption of screening algorithms.
How predictive analytics can help find the rare disease patient
- 1. PAGE 38 IMS HEALTH REAL-WORLD EVIDENCE SOLUTIONS INSIGHTS COMMERCIAL AND MARKET ACCESS John Rigg, PHD Principal and Head of Predictive Analytics, RWE Solutions, IMS Health John.Rigg@uk.imshealth.com How predictive analytics can help find the rare disease patient Early and accurate diagnosis of diseases is key to optimizing outcomes but particularly challenging in rare disorders, which often go undetected for years. new tools leveraging real-world data and innovation in advanced analytics are creating opportunities for dramatic improvements in identifying hard-to-find patients. Pioneering examples in studies of a rare multi-system disease and a cardiovascular condition demonstrate exciting potential, signaling a role for their use in broader strategies to accelerate effective treatment.
- 2. ACCESSPOINT • VOLUME 6 • ISSUE 11 PAGE 39 Although exact definitions vary, rare diseases are those affecting a small percentage of people.1 In the EU this is considered to mean no more than 5 cases per 10,000 individuals, and in the USA fewer than 200,000 individuals at any one time. But with somewhere in the region of 7,000 rare diseases already identified, collectively their burden is considerable: there are now an estimated 350 million sufferers worldwide – more than AIDS and cancer put together.2 And with, on average, five new rare diseases being described in the medical literature each week,3 patient numbers continue to grow. Often genetic and frequently chronic, life-threatening and debilitating,1,3 most rare diseases cannot be cured. However, significant developments in precision medicinei have seen notable breakthroughs in recent years, including highly targeted therapies addressing causal factors rather than symptoms alone. With more opportunities for treatment and over 450 orphan drugsii in development, the outlook for many individuals is increasingly hopeful.4 Yet even as progress continues, patients face tremendous barriers in benefiting from these innovations. In particular they struggle to obtain a timely and accurate diagnosis, which is difficult to deliver for the diverse range of rare disorders with widely varying signs and symptoms.1 Delays arise from misdiagnoses, multiple consultations and administration of inappropriate interventions (Figure 1). These delays all too often leave rare diseases undetected until a stage when even exceptional treatments are less effective. Given the progressive nature of many rare diseases, the consequences of late diagnosis can be immense and include physical deterioration, unnecessary stress and in some cases death.8 There are also significant financial implications, both for the individuals involved and society as a whole, reflecting the multiplicity of physician visits and greater use of costly diagnostic procedures.9 Furthermore, as observed by Kole and Faurisson,10 late- stage diagnosis can impede knowledge building in rare diseases, preventing improved understanding of the early manifestations of particular conditions. As a key element of broader efforts to improve awareness and management of rare diseases, solutions to enable earlier detection are imperative. In the words of one patient who experienced a five-year delay in being diagnosed with sarcoidosis: “As for advice about getting a diagnosis, I wish I had some magic formula for others.”11 In fact, the opportunity to find one has never been better. continued on next page Identifying patients to accelerate treatment Diagnosis for Rare Diseases Based on a European survey covering 8 rare diseases5 are initially misdiagnosed. This has resulted in over 40% of patients Patients visit an average of 7.3 physicians prior to an accurate diagnosis.6 5.6 years6 7.6 years A patient with a rare disease to receive the correct diagnosis waits on average 30%of patients have received three or more misdiagnoses7 1 out of 6 patients going through surgery and 1 out of 10 patients receiving psychiatric treatment Figure 1: Delays in diagnosis for rare diseases i The branch of medicine that enables healthcare services and treatment tailored to the specific genetic makeup of the individual (IMS Health RWE Dictionary; http://rwedictionary.com/) iI Medicinal products intended for diagnosis, prevention or treatment of life-threatening or debilitating rare diseases (http://www.eurordis.org/about-orphan-drugs)
- 3. PAGE 40 IMS HEALTH REAL-WORLD EVIDENCE SOLUTIONS INSIGHTS COMMERCIAL AND MARKET ACCESS a groundbreaking, real-world approach Alongside the science that is revolutionizing the treatment of rare diseases are two broader, parallel developments: the expansion of real-world data (RWD) and innovation in the analytical methods that can be used to interrogate the data. When brought together and supported by clinical insight these allow for the development of screening algorithms, presenting a major opportunity to help detect new, undiagnosed patients. 1. Expansion of RWD: The exponential increase in electronic healthcare information includes rich data from sources such as primary care Electronic Medical Records (EMRs), hospitals, claims information and patient registries. Detailed RWD on symptomology, diagnoses, consultations, treatment history, lab tests, etc, is routinely collected from patients and anonymized. Coupled with the growing ability to integrate data from different sources, there is now an unparalleled data foundation for finding undiagnosed patients. 2. Innovation in analytics: The field of advanced analytics has evolved greatly in industries such as finance and consumer goods. Sophisticated predictive techniques and algorithms that revolutionized facial recognition systems in on-line search engines are now helping to solve complex problems in healthcare. Machine learningiii technology, for example, can be used to identify complex, subtle patterns in the data of diagnosed patients to assist in detecting new sufferers of a disease. Pattern recognition techniques leveraging machine learning have successfully found ‘the needle in the haystack’ of undiagnosed patients with rare conditions. Demonstrating dramatic improvements in detection IMS Health has conducted pioneering research which demonstrates the power of RWD in helping to solve the problem of under diagnosis in rare diseases. This is illustrated in the following two recent case studies: the first shows how potentially undiagnosed patients can be found using screening tools based on advanced analytics; the second how RWD can be used to identify potential health system barriers to diagnosis. Both approaches are complementary, tackling patient-level and system-wide challenges respectively. Case study 1 Finding patients at high risk of a rare multi-system disease earlier Highly promising results from a recent study in a rare multi-system disease demonstrate the application of a screening algorithm in the UK. Specifically, the focus was to detect patients with this particular condition, which is substantially under diagnosed and where up to 40% of identified patients are diagnosed late, often by decades. Potentially undiagnosed patients were identified from routinely collected primary care and administrative data by virtue of advanced machine learning methods incorporating clinical expertise which revealed patterns in the data that were predictive of disease presence. The analysis was conducted in two stages leveraging de- identified EMRs. Firstly, analytics experts developed a screening algorithm using classical statistical methods combined with clinical expertise. Secondly, they applied advanced machine learning methods to refine and optimize the algorithm. A test (‘blind’) sample of 70,000 randomly selected patients was risk-scored by the initial algorithm without knowing which patients had a confirmed diagnosis for the disease. This produced a high-risk group containing 8% of confirmed cases. The test sample was then risk scored by the refined algorithm, exploiting machine learning techniques. This produced a prevalence of the confirmed diagnosis in the highest risk group of 20.5%. Given that only 0.7% of patients in the test sample actually had the disease, the evidence suggests that the algorithm could be used to dramatically increase the odds of finding high-risk patients earlier (Figure 2). Confirmed Diagnosis Potentially Undiagnosed Candidates For Screening Potential Undiagnosed Potential Diagnosis via Algorithm Diagnosed Cases 50 50 39 84 45 Source: IMS Health Figure 2: Potentially undiagnosed patients identified for screening 100 80 60 20 0 40 120 140 160 180 200 220 240 NumberofPatients Center Carrying Out Diagnostic Procedure Source: IMS Health Tertiary Center 1 Tertiary Center 2 Tertiary Center 3 Tertiary Center 4 Tertiary Center 5 Other 230 148 72 70 62 39 Number of Patients Diagnosed by Center Figure 3: Variation in disease incidence by tertiary center iii A collection of advanced, data-driven statistical methods which can be used to identify complex patterns in data.
- 4. ACCESSPOINT • VOLUME 6 • ISSUE 11 PAGE 41 Case study 2 Identifying health system barriers causing under diagnosis A second case study is an analysis which was conducted to determine, in a complex, multi-center diagnosis pathway, whether a lengthy diagnosis process could be a causal factor in late presentations of a potentially fatal rare cardiac disease. If detected early, the condition could be reversible or manageable with treatment. In a process involving literature and data profiling, a cohort selection algorithm was developed leveraging secondary care data in a major EU country. The selected cohort triangulated well with literature incidence and demographic values and enabled health service usage and diagnosis patterns to be investigated for a period of more than five years (April 2009 to October 2014). The analysis revealed a high number of events (21 for the average patient) for three years ahead of a formal diagnosis, with over 90% of patients being known to the hospital system within the three-year time frame. It also demonstrated wide variation in the types of diagnostic pathways followed to reach a tertiary center initially. Patients were found to see on average three different hospital centers in the three years pre-diagnosis and five different specialty types. Furthermore, the study identified substantial variability in the incidence rate per 100K population, being higher in regions feeding in to the leading diagnosis center (Figure 3). This suggested challenges of under diagnosis in other parts of the country. Insights from this research were positively received by leading clinical experts in the field as a novel and previously unseen perspective on their patient population. The study generated hypotheses for further work and served as a basis for building a rich pool of RWD to inform this therapy area, in association with academic and clinical institutions. These studies illustrate the power of techniques enabled by the use of RWD and analytics to facilitate broader efforts to reach patients suffering from a rare disease with treatment that could, potentially, be curative. Impressive results but challenges remain Evidence now exists to show how the application of advanced analytics to large-scale RWD can help identify undiagnosed patients with rare diseases. These screening algorithms can form an important part of the portfolio of strategies to bring the right treatment, including innovative new therapies, to patients with rare diseases. The initial results are impressive: the growing availability of RWD creates a rich foundation for screening algorithms while developments in advanced machine learning and predictive analytics, such as signal detection theory, enable the distinction between ‘signal’ and ‘noise’ to make algorithms accurate and cost-effective. However, there are challenges to address before they can be employed to flag high-risk patients from medical records alone: patient confidentiality has to be protected; underlying data must be sufficiently broad to reach a critical mass of these hard-to- find patients; and clinicians must be willing to embrace results from screening algorithms. Nevertheless, if there is serious intent to develop treatments to improve the lives of patients with rare diseases, then there should be equally serious efforts to find those patients. RWD and predictive analytics can undoubtedly play an important role in helping to achieve this goal. 1 What is a rare disease? EURORDIS. Rare Diseases Europe. http://www.eurordis.org/content/what-rare-disease 2 Global Genes. Available at: https://globalgenes.org/who-we-are-2/ Accessed 6 Dec 2015 3 Medicines for rare diseases. European Medicines Agency. http://www.ema.europa.eu/ema/index.jsp?curl=pages/special_topics/general/general_content_000034.jsp 4 PhRMA. A Decade Of Innovation in Rare Diseases 2005-2015. PhRMA, 2015. Available at: http://www.phrma.org/sites/default/files/pdf/PhRMA-Decade-of-Innovation-Rare-Diseases.pdf Accessed 6 December 2015 5 EURORDIS – Rare Diseases Europe. Survey of the delay in diagnosis for 8 rare diseases in Europe (‘EURORDISCARE2’). Available at: www.eurordis.org/IMG/pdf/Fact_Sheet_Eurordiscare2.pdf 6 Engel PA, Bagal S, Broback M, Boice N. Physician and patient perceptions regarding physician training in rare diseases: The need for stronger educational initiatives for physicians. Journal of Rare Disorders, 2013; 1(2): 1-15. http://www.journalofraredisorders.com/pub/IssuePDFs/Engel.pdf 7 Limb L, Nutt S, Sen A. Experiences of rare diseases: An insight from patients and families. Rare Disease UK. December, 2010. Available at: http://www.raredisease.org.uk/documents/RDUK-Family-Report.pdf Accessed 6 December, 2015 8 EURORDIS. Voice of 12,000 patients. Experiences and Expectations of Rare Disease Patients on Diagnosis and Care in Europe. A report based on the EurordisCare2 and EurordisCare3 Surveys. EURORDIS, 2009. Available at: http://www.eurordis.org/publication/voice-12000-patients 9 Rare Disease Impact Report: Insights from patients and the medical community. Shire, April 2013. Available at: http://www.geneticalliance.org.uk/docs/e-update/rare-disease-impact-report.pdf and https://www.shire.com/newsroom/2013/april/shire- launches-report Accessed 6 December, 2015 10 Kole A, Faurisson F. Rare diseases social epidemiology: Analysis of inequalities. Advances in Experimental Medicine and Biology, 2010; 686: 223-50 11 Inspire. The road to diagnosis: Stories from patients with rare diseases. 2011. https://www.inspire.com/static/inspire/reports/inspire- rare-disease-day-report-2011.pdf