Assessing Decline: Visualising Progression in Huntington’s Disease using a Clinical Dashboard with Enroll-HD Data

Abstract

Background: In Huntington’s disease (HD), it remains unclear how symptom severity and rate of symptomatic change relates to age and CAG repeat number (CAGn). It is often difficult for clinicians to assess whether an affected individual’s symptoms are progressing at a similar rate to their affected peers, limiting their ability to intervene at the most appropriate time.

Objective: To develop a clinical dashboard that compares an individual’s total motor score (TMS), total functional capacity (TFC) and symbol digit modality test (SDMT) scores against a global cohort, controlling for age and CAGn. The dashboard could then be used by clinicians to identify individuals progressing at a disproportionate rate to his or her peers.

Methods: Annualised longitudinal clinical assessment scores from the Enroll-HD dataset were used to generate decline trajectories of the global cohort, allowing cross-sectional (TMS n = 734; TFC n = 734; SDMT n = 694) and longitudinal (TMS n = 270; TFC n = 270; SDMT n = 247) comparison with individual clinical symptom rating scores, to assess decline relative to affected peers.

Results: An electronic dashboard with a dynamic output display was created that rapidly compares clinical symptom rating scores of a specific individual against affected peers from a global cohort of comparable CAGn.

Conclusions: This study shows the potential for use of multi-centre trial data in allowing comparison of the individual to a larger group to facilitate improved decision-making for individual patients. Visualisation of these metrics via a clinical dashboard demonstrates how it may aid identification of those with disproportionate decline, offering potential for intervention at specific critical points in the disease course.

Keywords

Huntington’s disease clinical dashboard longitudinal Enroll-HD

INTRODUCTION

A clinical dashboard is a “method of visually displaying aspects of performance data relevant to patient care” [1]. Dashboards typically integrate large amounts of numerical information, and assimilate these into a visually-powerful interface, which can then be interpreted by clinical staff to inform and influence clinical practice [1]. They have been embraced across a range of healthcare specialties such as emergency care [2, 3], neonatal intensive care [4], and maternity [5] – and were noted by NHS England as a powerful way of “inform[ing] the daily decisions that drive quality improvement” [6].

Computers perform well in fields that human brains may struggle, such as in rapidly processing large amounts of disparate numerical data [7]. Consequently, there is huge potential for dashboards to complement clinician expertise in developing a holistic approach to informing care. Despite this, the use of a Huntington’s disease clinical dashboard has not previously been reported in the literature.

Access to reliable data is an important factor that shapes the quality of tools such as clinical dashboards [8], and the development of international multi-centre HD data collection, via studies such as Enroll-HD, offers great potential. In HD, the clinical manifestations of progressive motor dysfunction, cognitive deterioration and psychiatric disturbance [9] occur in patients with at least 35 Huntingtin (HTT) gene CAG repeats (CAGn ≥35) [10]. The ability to quantify this in every individual, and the systematic recording of symptomatic changes, using the Unified Huntington’s Disease Rating Scale (UHDRS) [11], facilitates the comparison of the individual against the group.

We note that it remains challenging for a single clinician to assess whether symptoms of HD in an individual are progressing at a typical rate for someone of that demographic. This uncertainty is compounded by conflicting research data, some of which conclude that there is a link between CAGn and clinical HD progression [12 –14], and others suggesting there is an ambiguous association [15, 16] or none at all [17].

We therefore postulated that pooling the longitudinal multi-centre trial data from Enroll-HD would provide a foundation to develop a clinical dashboard that compares, and visually represents, the symptomatic progression of an HD-manifest patient with a cohort of similar peers. The development of such a tool could help clinicians rapidly identify whether a specific patient is experiencing disproportionate decline compared to that predicted from group data, and could be used alongside a clinician’s judgement as a triaging tool to suggest specific intervention points.

Our aim was to develop a dashboard that was tailored to the individual, and required minimal data entry, to represent the comparison outputs immediately [18] and in a visually-powerful manner [19]. These features have been previously recognised [18, 19] as highly valued features of innovative tools, and our intention was to incorporate these elements into the dashboard design.

METHODS

Data from 745 motor-manifest participants in the Enroll-HD study [20] were used, from the January 2015 data cut. All participants were assessed using the UHDRS and the data were passed through quality control and de-identification processes [20]. Figure 1 illustrates the selection criteria for the participants, including that patients must be manifest, have a specific age recorded, and have clinical assessment scores available, either for baseline only (cross-sectional part of dashboard), or for both baseline and follow-up (longitudinal part of dashboard).

Fig.1

Flow diagram showing the recruitment of participants to the dashboard. The process undertaken by Enroll-HD (adapted from Enroll-HD Data Handling Manual [20]) and the process undertaken in this publication have been distinguished.

CAGn was selected as an important demographic variable due both to its role in determining age of symptomatic onset [10] and also its possible association with clinical HD progression [12 –16]. Once participants’ CAGn was identified, they were then allocated one of five CAGn bands (a CAGn of ≤41, 42–43, 44–45, 46–48, ≥49), depending on their CAGn. These CAGn bands were used instead of absolute CAGn values to provide sufficient data points in each group for a trend to emerge. With the expansion of the data source and more individuals present for each CAGn, over time, these bands can be eliminated and replaced with regression lines grouped by absolute CAGn values.

Total motor score (TMS), total functional capacity (TFC) and symbol digit modality test (SDMT) were selected as the key UHDRS clinical assessment scores. TMS, where an increasing score indicates a higher burden of motor pathology [21], defines clinical diagnosis [22], is widely used [22], and is well validated as a marker of clinical progression, particularly following diagnosis [11 , 23]. TFC, where a low score indicates worse functional ability [11, 21], is an important outcome measure in clinical trials [24]. The SDMT, where a low score indicates worse performance [21], is very sensitive for follow-up of HD patients [25] and is also widely-used.

The demographic breakdown (age, CAGn, sex and geographical distribution) of participants is provided by Table 1.

Table 1

Demographic information of participants

Dashboard	UHDRS	n	Mean	Mean	% females	% of participants from each region (n)
component	component		age (SD)	CAGn (SD)	(n)	North	Europe	Oceania	Latin
						America			America
Cross-sectional	TMS	734	52.6 (12.2)	43.8 (3.0)	51 (376)	56 (410)	36 (265)	6 (46)	2 (13)
	TFC	734	52.6 (12.2)	43.8 (3.0)	51 (376)	56 (410)	36 (265)	6 (46)	2 (13)
	SDMT	694	52.5 (12.2)	43.7 (3.0)	50 (350)	56 (389)	36 (248)	6 (45)	2 (12)
Longitudinal	TMS	270	51.6 (12.1)	43.7 (2.9)	50 (136)	94 (255)	0 (1)	3 (7)	3 (7)
	TFC	270	51.6 (12.1)	43.7 (2.9)	50 (136)	94 (255)	0 (1)	3 (7)	3 (7)
	SDMT	247	51.6 (12.1)	43.7 (2.9)	49 (122)	94 (232)	0 (1)	3 (7)	3 (7)

Three cross-sectional, and three longitudinal, regression equations were defined for later use in the output display of the dashboard. The three cross-sectional regressions used raw TMS, TFC or SDMT score as the dependent variable, and age of participant at baseline as the independent variable. In the three longitudinal regression equations, annualised TMS, TFC or SDMT change over the year prior to consultation was defined as the dependent variable, and age of participant at follow-up as the independent variable.

The independent variables of the longitudinal regression slopes were calculated using annualisation, a previously-described technique [26]: baseline UHDRS clinical assessment scores were subtracted from the follow-up score, then divided by the number of days between these two recordings, and multiplied by 365. Consequently, all scores were standardised as the amount of change in 365 days prior to the consultation. Scores were rounded to the nearest integer. Each participant’s age at follow-up was calculated as age at baseline plus one year. Longitudinal calculations included only participant data from baseline and one-year follow-up. This is because the small amount of data from two- and three-year follow ups in the January 2015 Enroll-HD data cut meant that generating a meaningful regression was impractical with such sporadic data.

To assess whether bias had been introduced by participants not continuing to follow-up, independent-samples t-tests were conducted, comparing those who continued to follow-up and those who did not for age and CAGn.

The dashboard was developed using Microsoft Excel 2016 with the aim of graphic comparison of an individual patient’s key UHDRS clinical assessment scores with those of Enroll-HD participants in the same CAGn band. The input section of the dashboard required a specific patient’s CAGn, date of birth and date of consultation (demographic information), along with TMS, TFC and SDMT scores from each consultation.

For each cross-sectional longitudinal regression slope, the independent and dependant variables were plotted in the dashboard and grouped based on CAGn band. Regression lines of slope y = mx+c (Tables 2 and 3) were fitted, along with 95% confidence and 95% prediction intervals, to which polynomial best fit lines were added. Specific data points were then hidden, showing only the lines. The dashboard was programmed to display only the regression line of the CAGn band that matched the patient’s input values. Visual Basic for Applications (VBA) code was created and assigned to a button which, when selected, plotted a patient’s cross-sectional and longitudinal scores onto the regression lines.

Table 2

Descriptive statistics of the cross-sectional regression slopes

CAG band	No. of participants	Regression equation	R²	Standard error
Total Motor Score
≤41	150	y = 0.684x – 10.8	0.121	16.1
42–43	255	y = 1.08x – 26.2	0.224	16.2
44–45	150	y = 1.38x – 31.2	0.259	16.9
46–48	107	y = 1.86x – 37.8	0.417	15.9
≥49	72	y = 0.67x + 18.0	0.036	20.4
Total Functional Capacity
≤41	150	y = –0.0970x + 14.9	0.063	3.25
42–43	255	y = –0.160x + 17.5	0.149	3.09
44–45	150	y = –0.215x + 18.7	0.190	3.21
46–48	107	y = –0.260x + 18.7	0.269	3.10
≥49	72	y = –0.0524x + 9.49	0.007	3.66
Symbol Digit Modality Test
≤41	146	y = –0.637x + 66.8	0.186	11.7
42–43	241	y = –0.671x + 63.2	0.194	10.9
44–45	142	y = –0.990x + 71.9	0.326	10.4
46–48	102	y = –0.881x + 60.1	0.162	13.8
≥49	63	y = –0.507x + 40.6	0.045	13.6

Table 3

Descriptive statistics of the longitudinal regression slopes

CAG band	No. of participants	Regression equation	R²	Standard error
Total Motor Score
≤41	49	y = 0.0981x –4.42	0.011	8.49
42–43	98	y = –0.000182x + 2.05	<0.001	7.84
44–45	60	y = 0.154x – 3.13	0.018	8.31
46–48	37	y = –0.0502x + 6.30	0.001	7.25
≥49	26	y = 0.0804x + 2.57	0.003	9.73
Total Functional Capacity
≤41	49	y = –0.0167x + 0.568	0.018	1.13
42–43	98	y = 0.00105x – 0.833	<0.001	1.69
44–45	60	y = –0.0377x + 1.45	0.015	2.25
46–48	37	y = –0.0363x + 1.30	0.010	1.92
≥49	26	y = –0.0109x –0.564	0.001	1.91
Symbol Digit Modality Test
≤41	44	y = 0.114x –9.83	0.038	5.04
42–43	94	y = –0.0101x –0.981	<0.001	4.98
44–45	54	y = –0.0644x –0.242	0.006	6.44
46–48	34	y = 0.132x –7.86	0.013	6.10
≥49	21	y = –0.368x + 9.65	0.040	10.6

Further VBA code was created and assigned to a button that copies past clinical assessment scores into a ‘Past record’ section, which stores every score previously entered for that patient. VBA code was also created and assigned to the RESET button, which, when selected allows all patient data to be instantly cleared, ready for a new patient.

All data were provided for this project anonymously from REC-approved sites participating in the Enroll-HD Study.

RESULTS

Annual follow up consultations were completed at a mean interval of 381 days, with a standard deviation of 49 days and a range of 791 to 285 days (n = 270). There was no significant difference in age (mean 51.6 vs 53.2 years; p = 0.09) or CAGn (mean 43.7 vs 43.8 repeats; p = 0.7) for those who continued with follow up and those who did not.

The dashboard used the CAGn regression data as previously described in the Methods section. Figure 2 shows the full dashboard, which, to demonstrate its functionality, includes data from three consultations for an anonymised Enroll-HD participant. The input screen, which is visible in the top left corner of the dashboard, has been designed for ease-of use [18]. Figure 3 demonstrates the dynamic nature of the dashboard, where the output is dependent upon the CAGn entered onto the patient’s input screen. This shows how longitudinal outcomes are highly dependent upon CAGn, and offer visualisation of the potential relevance of an individual’s actual performance in relation to their predicted ability.

Fig.2

Screenshot of the populated dashboard. On the input screen (top left corner), data input into the “CAGn” cell from the “Demographic information” section informs which trend lines to display on the graphs. “DOB” is used to calculate the patient’s age at each consultation. “TMS”, “TFC” and “SDMT” cells on the “Most recent consultation” section are the scores recorded in clinic. The date in this section is used alongside DOB to calculate age at consultation. Button 1 links to VBA code that clears the “Most recent consultation data” and stores it in the “Past record” section, thus allowing new consultation data to be entered. Button 2 links to VBA code that plots cross-sectional and longitudinal data onto the graph. “RESET” links to VBA code that clears all data, thus resetting the system. The purple box automatically updates to remind the user what CAGn band the patient belongs to. Each black shape (diamond, triangle, square) represents patient clinical assessment scores from separate consultations, with the square representing the most recent consultation data. The ‘Past record’ table expands to the right with each subsequent consultation. On the output screen, the patient’s cross-sectional (top graphs) and longitudinal (bottom graphs) data are visualised, showing regression lines fitted to each CAGn band (unbroken straight line). 95% prediction intervals are the outermost lines with longer dashes. 95% confidence intervals of the best fit line are the dashed lines closest to the unbroken best fit line. The bottom graphs will always have one fewer data points compared to the top graphs because at least two consultations are required to make a longitudinal assessment.

Fig.3

Left: graphical display if the patient’s CAGn is 45. Right: graphical display if the patient’s CAGn is 48. The black shapes (patient clinical assessment score / change in score) have the same values on the left and right. Once a patient’s demographic data are entered, regression lines of Enroll-HD data relevant only to the patient’s CAGn band appear on the graphs. Consequently, the individual is only being compared to Enroll-HD participants who share a similar demographic.

DISCUSSION

In HD, assessment of the patient’s journey involves collection and assimilation of information from a range of clinicians [27] over a prolonged period of time. Truly comprehending the long-term significance of these data, where there is such great symptomatic variation, is a major challenge. Monitoring of symptoms through clinical dashboards has great potential for powerfully integrating these data, informing clinicians and improving outcomes [18]. This work developed a dashboard that offers support for clinicians in building an overall picture of a patient’s clinical status relative to similarly assessed peers, and shows the practical contribution that large multi-centre data repositories such as Enroll-HD can provide to the care of the individualpatient.

The six regressions encompassing the output display demonstrate how the cross-sectional and longitudinal components of the dashboard can complement each other to build an overall clinical picture for the patient, as has been noted in other areas of HD research [28]. The three cross-sectional graphical displays illustrate the patient’s absolute clinical assessment scores compared to his or her peers, whilst the longitudinal displays compare the rates of change. This asset has potential to provide real clinical benefit. For example, a rapidly-declining patient may experience a noticeable improvement in longitudinal score following initiation of medical treatment, while still scoring relatively poorly on clinical assessment scores within the cross-sectional displays. This longitudinal improvement could offer the basis of an encouraging discussion in clinic. The use of clinical dashboards in such a manner could improve, and even standardise, aspects of care and care protocols by providing an objective marker for specific intervention points. The pooled Enroll-HD dataset used in this dashboard encompasses a large experience of HD clinical assessment, thereby facilitating benchmarking against a larger cohort of equivalent peers than is available to an individual clinician at aspecific site.

Graphical representation of patient data is a well-established way of improving decision making and positively improving healthcare standards [18], and this was an attribute we aimed to capture within this dashboard. The clean input screen and rapidly-updating displays (through adopting VBA codes and formulae) were designed with the appreciation that the aid would be unlikely to be embraced if information retrieval takes more than about 30 seconds [18]. These factors all shaped its design as a tool which immediately visualises individual performance, and helps with identification of those most at risk of poor outcome, providing clinicians with the potential to intervene early.

LIMITATIONS

There are some limitations to this work which we regard as proof-of-concept and first-of-a-kind in HD. Firstly, we plotted regression lines based on CAGn bands, rather than absolute CAGn values. It would be preferable to group the regression lines based on absolute CAGn values, but by grouping into CAGn bands, it ensured that enough data points were present in each group to generate a strong trend. As the data source is further enriched with new individuals, these CAGn bands will be able to be removed to allow regression lines to be grouped based on absolute CAGn values.

Furthermore, the width of the 95% prediction intervals illustrates a limitation of the dashboard’s ability to identify outliers. An attempt to reduce this interval size in TMS was made by identifying dystonia [28, 29] and chorea [30] as those aspects suffering particularly from inter-rater and intra-rater variability, and eliminating them from the TMS. However, this proved on qualitative assessment to make a negligible difference to the size of the prediction intervals, and so the attempt to remove chorea and dystonia from TMS was abandoned.

The low R² values of the linear regression equations, as shown in Tables 2 and 3, highlights a limitation in the extent to which these regressions fully capture the dynamic nature of HD as a complex neurodegenerative condition. Consequently, the future identification of additional input variables –beyond CAGn and age –could help to tailor the output more appropriately to the patient in clinic, thus improving its inherent predictive value.

There was a potential risk of annualisation exaggerating small changes in clinical assessment scores if two consultations occurred close together. However, no follow-up in the Enroll-HD data occurred earlier than 294 days after baseline (as per Enroll-HD protocol for annual visits [20]), a sufficient time for annualisation to be largely representative of the overall rate of change. Therefore, when using the dashboard, it is recommended that patients’ longitudinal graphs are only considered if follow up occurs at least 294 days (or approximately 9 months) after baseline (the lower limit of the range used in the Enroll-HD data). Age, rather than time since diagnosis, was selected as the independent variable on the graphs as objective identification of an absolute time of diagnosis has been recognised as unreliable [31]. On balance, we concluded that absolute age would be the most reliable independent variable.

Finally, it should be noted that North American participants contributed to 56% of the cross-sectional data, and 94% of the longitudinal data (Table 1), creating the potential for bias in the regression equations used in the dashboard. Over time, with expansion of the number of observations in the Enroll-HD dataset available for modelling, this is likely to become less relevant.

FUTURE DIRECTIONS

As research into the symptomatic progression of HD strengthens, so will the understanding of genetic and non-genetic influencers of long-term clinical outcomes. This clinical dashboard could be advanced by introducing new variables to the input screen (alongside the current variables of CAGn and date of birth), in order to allow for more tailored information to be displayed. For example, body mass index (where a small observational study has previously associated a higher score with slower symptomatic progression [33]), or concomitant medications (some of which have been associated with increased rates of motor decline [33]) could feasibly be included as variables if indicated as significant factors in symptomatic progression. Meanwhile, recent work into potential genetic modifiers of HD –such as the role specific loci [34] and polymorphisms [35] –may lead to a more comprehensive genetic picture of HD, which could also be built into the input screen of the dashboard, and the subsequent regression calculations. Further stratification into predominantly choreiform versus hypokinetic-rigid disease may help further refine progression norms for the individual compared with the group.

Future implementation of an input screen which more accurately reflects the patient’s complex clinical history should go some way to countering these limitations. However, it is important to recognise that as the output becomes more tailored, fewer relevant data points will contribute to the graphs, and so continually adding comprehensive source data from Enroll-HD should be a priority.

The implementation of an online clinical dashboard with real-time updates of the source data would also be a significant advance –an idea which has already been postulated in other fields [36]. This would provide a constantly-expanding pool of source data, which should remove the requirement for CAGn bands, while also possibly shrinking the prediction intervals. Such a feature could be provided within the Enroll-HD platform, and provide a direct, and immediate, clinical benefit to research participation.

CONCLUSION

The accumulation of multi-trial longitudinal data through the Enroll-HD study affords an excellent opportunity for further development of the clinical dashboard concept. This work has demonstrated a powerful way of utilising such data to aid clinicians’ evaluation of an individual’s symptomatic progression in HD. While dashboards should be used holistically alongside –rather than instead of –a clinician’s expertise, this dashboard’s ability to rapidly compare a specific patient in clinic with global peers makes it a highly valuable tool for flagging disproportionate clinical decline. This would then prompt review to determine the causes, and potential reversibility, of such decline. Conversely, it may be beneficial in identifying individuals progressing at a slower rate than expected, providing encouragement, and possibly offering support for any interventions (pharmacological and non-pharmacological alike). Accurately modelling chronic degenerative diseases such as HD is highly complex, and this dashboard’s effectiveness in improving clinical care is yet to be determined. However, the success of dashboards in other fields of medicine makes us optimistic that this concept, once adopted, will also improve patient outcomes.

CONFLICTS OF INTEREST

The authors have no conflict of interest to report.

Footnotes

ACKNOWLEDGMENTS

CMK received support from the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care (CLAHRC) Wessex. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.

Enroll-HD is a longitudinal observational study for Huntington’s disease families intended to accelerate progress towards therapeutics; it is sponsored by CHDI Foundation, a nonprofit biomedical research organization exclusively dedicated to developing therapeutics for HD. Enroll-HD would not be possible without the vital contribution of the research participants and their families.

References

Phull

, Hall

. Clinical dashboards and their use in an adult mental health inpatient setting, a pilot study. Clin Gov. 2015;20(4):199–207.

Swartz

, Cimino

, Fred

, Green

, Vawdrey

. Designing a clinical dashboard to fill information gaps in the emergency department. AMIA Annu Symp Proc. 2014;2014:1098–104.

Staib

, Sullivan

, Jones

, Griffin

, Bell

, Scott

. The ED-inpatient dashboard: Uniting emergency and inpatient clinicians to improve the efficiency and quality of care for patients requiring emergency admission to hospital. Emerg Med Australas. 2016. doi: 10.1111/1742-6723.12661

Raban

, Bamford

, Joolay

, Harrison

. Impact of an educational intervention and clinical performance dashboard on neonatal bloodstream infections. S Afr Med J. 2015;105(7):564–6.

Chandraharan

, Arulkumaran

. The role of clinical dashboards in improving patient care: Experience with the ‘Maternity Dashboard’. Ceylon Med J. 2016;61(2):83–5.

Department of Health (DoH). “High quality care for all”, NHS Next Stage Review final report, DoH, London. 2008.

Johnson

. Designing with the Mind in Mind: Simple Guide to Understanding User Interface Design Rules. Massachusetts, USAMorgan Kaufmann; 2010.

Chandraharan

. Clinical dashboards: Do they actually work in practice? Three-year experience with the Maternity Dashboard. Clin Risk. 2010;16(5):176–82.

Tabrizi

, Langbehn

, Leavitt

, Roos

, Durr

, Craufurd

, et al. Biological and clinical manifestations of Huntington’s disease in the longitudinal TRACK-HD study: Cross-sectional analysis of baseline data. Lancet Neurol. 2009;8(9):791–801.

10.

Andrew

, Goldberg

, Kremer

, Telenius

, Theilmann

, Adam

, et al. The relationship between trinucleotide (CAG) repeat length and clinical features of Huntington’s disease. Nat Genet. 1993;4, 398–403.

11.

Beglinger

, Duff

, Allison

, Theriault

, O’Rourke

, Leserman

, et al. Cognitive change in patients with Huntington disease on the Repeatable Battery for the Assessment of Neuropsychological Status. J Clin Exp Neuropsychol. 2010;32(6):573–8.

12.

Kuan

, Kasis

, Yuan

, Mason

, Lazar

, Barker

, et al. Modelling the natural history of Huntington’s disease progression. J Neurol Neurosurg Psychiatry. 2015;86(10):1143–9.

13.

Brandt

, Bylsma

, Gross

, Stine

, Ranen

, Ross

. Trinucleotide repeat length and clinical progression in Huntington’s disease. Neurology. 1996;46(2):527–31.

14.

Ravina

, Romer

, Constantinescu

, Biglan

, Brocht

, Kieburtz

, et al. The relationship between CAG repeat length and clinical progression in Huntington’s disease. Mov Disord. 2008;23(9):1223–7.

15.

Rosenblatt

, Liang

, Zhou

, Abbott

, Gourley

, Margolis

, et al. The association of CAG repeat length with clinical progression in Huntington disease. Neurology. 2006;66(7):1016–20.

16.

Mahant

, McCusker

, Byth

, Graham

, Group

. Huntington’s disease: Clinical correlates of disability and progression. Neurology. 2003;61(8):1085–92.

17.

Kieburtz

, MacDonald

, Shih

, Feigin

, Steinberg

, Bordwell

, et al. Trinucleotide repeat length and progression of illnessin Huntington’s disease. J Med Genet. 1994;31(11):872–4.

18.

Badgeley

, Shameer

, Glicksberg

, Tomlinson

, Levin

, McCormick

, et al. EHDViz: Clinicaldashboard development using open-source technologies. BMJOpen. 2016;6(3):e010579.

19.

Mould

, Upton

, Wojciechowski

. Dashboard systems:Implementing pharmacometrics from bench to bedside. AAPS J. 2014;16(5):925–37.

20.

The Enroll-HD Study Group. Enroll-HD: A prospective registry study in a global Huntington’s disease cohort. A CHDI Foundation Project. Enroll-HD Data Handling Manual, Version 2015-01-R1. 2015.

21.

Unified Huntington’s Disease Rating Scale: Reliability and consistency. Mov Disord. 1996;11(2):136–42.

22.

Long

, Paulsen

, Marder

, Zhang

, Kim

, Mills

, et al. Tracking motor impairments in the progression of Huntington’s disease. Mov Disord. 2014;29(3):311–9.

23.

Kirkwood

, Su

, Conneally

, Foroud

. Progression of symptoms in the early and middle stages of Huntington disease. Arch Neurol. 2001;58(2):273–8.

24.

Dorsey

, Beck

, Darwin

, Nichols

, Brocht

, Biglan

, et al. Natural history of Huntington disease. JAMA Neurol. 2013;70(12):1520–30.

25.

Lemiere

, Decruyenaere

, Evers-Kiebooms

, Vandenbussche

, Dom

. Cognitive changes in patients with Huntington’s disease (HD) and asymptomatic carriers of the HD mutation–a longitudinal follow-up study. J Neurol. 2004;251(8):935–42.

26.

Ghosh

, Carpenter

, Rowe

. A longitudinal study of motor, oculomotor and cognitive function in progressive supranuclear palsy. PLoS One. 2013;8(9):e74486.

27.

El-Nimr

, Tams

. Working together: A one-stop shop for Huntington’s disease. Prog Neurol Psychiatry. 2011;15(1):15–7.

28.

Fox

, Brotchie

. Levodopa-induced dyskinesia in Parkinson’s disease. New York City, USA: Springer; 2014.

29.

Beghi

, Regio

, Papantonio

, Bentivoglio

, Fasano

, Fogli

, et al. Reliability of clinical diagnosis of dystonia. Neuroepidemiology. 2014;43(3-4):213–9.

30.

Novotný

, Rusz

, Čmejla

, Růžičková

, Klempíř

, Růžička

. Hypernasalityassociated with basal ganglia dysfunction: Evidence fromParkinson’s disease and Huntington’s disease. Peer J. 2016;4:e2530.

31.

McCusker

, Richards

, Sillence

, Wilson

, Trent

. Huntington’s disease: Neurological assessment of potential genecarriers presenting for predictive DNA testing. J Clin Neurosci. 2000;7(1):38–41.

32.

Myers

, Sax

, Koroshetz

, Mastromauro

, Cupples

, Kiely

, et al. Factors associated with slow progression in Huntington’s disease. Arch Neurol. 1991;48(8):800–4.

33.

Tedroff

, Waters

, Barker

, Roos

, Squitieri

, Group

ERS

. Antidopaminergic medication is associated with more rapidly progressive Huntington’s disease. J Huntingtons Dis. 2015;4(2):131–40.

34.

Consortium GMoHsDG-H. Identification of genetic factors that modify clinical onset of Huntington’s disease. Cell. 2015;162(3):516–26.

35.

de Diego-Balaguer

, Schramm

, Rebeix

, Dupoux

, Durr

, Brice

, et al. COMT Val158Met polymorphism modulates Huntington’s disease progression. PLoS One. 2016;11(9):e0161106.

36.

Felkey

, Fox

. Health System Dashboard: It’s all coming together. Hosp Pharm. 2014;49(5):485–6.