Brain and Cognition 56 (2004) 24–29 www.elsevier.com/locate/b&c
Variability in fundamental frequency during speech in prodromal and incipient Parkinson’s disease: A longitudinal case studyq Brian Harel,a,b Michael Cannizzaro,a and Peter J. Snydera,b,c,* a
c
Voice Acoustics Laboratory, Clinical Technology, Pfizer Global Research & Development Groton, CT, USA b Department of Psychology, University of Connecticut Storrs, CT, USA Departement de Psychologie, Centre de Neuroscience de la Cognition, Universite de Quebec a Montreal Montreal, Que., Canada Accepted 11 May 2004 Available online 9 June 2004
Abstract Nearly two centuries ago, Parkinson (1817) first observed that a particular pattern of speech changes occur in patients with idiopathic Parkinson’s disease (PD). Numerous studies have documented these changes using a wide variety of acoustic measures, and yet few studies have attempted to quantify any such changes longitudinally, through the early course of the disease. Moreover, no attempt has been made to determine if speech changes are evident during the prodromal period, prior to the onset of clinically noticeable symptoms. This case-control pilot study is a first attempt to determine if changes in fundamental frequency variability during speech, an acoustic measure known to be affected later in the course of the disease, are evident during the prodromal period. A retrospective analysis of videotape footage recorded and made available by a leading national television news service. Videotape samples were obtained for a single individual (and a well-matched control subject) over an 11-year period of this individual’s life (7 years prior to diagnosis of PD, and 3 years post-diagnosis). Results suggest that changes in F0 variability can be detected as early as 5 years prior to diagnosis (consistent with findings from other laboratories that have relied on cross-sectional study approaches). This pilot study supports the utility of such a design approach, and these results warrant continued effort to better understand the onset of PD and sensitivity of measurement of voice acoustical changes during the prodromal period. Ó 2004 Elsevier Inc. All rights reserved.
1. Introduction Idiopathic Parkinson’s disease (PD) is a progressive neurodegenerative disorder of unknown etiology that affects over 1 million people in North America (Lang & Lozano, 1998). The hallmark clinical symptoms of PD are largely motoric in nature and include a resting tremor, muscular rigidity, bradykinesia, akinesia, and postural abnormalities (Duffy, 1995). Patients do not usually become noticeably symptomatic until between the ages of 50 and 69 years (Hoehn & Yahr, 1967). Although no single causative factor for PD has been q The authors appreciate the advice and training provided by Dr. Henrı Cohen and his laboratory technician, Ga€elle Belanger. We also appreciate the kind assistance of Dr. Eve Pickering for her statistical advice. * Corresponding author. E-mail address:
[email protected]fizer.com (P.J. Snyder).
0278-2626/$ - see front matter Ó 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.bandc.2004.05.002
identified, several mechanisms are thought to be involved including mitochondrial defects, oxidative stress, glutamate toxicity, genetic factors, and apoptosis (Blandini, Giuseppe, Tassorelli, & Martignoni, 2000). Idiopathic PD differs from parkinsonism, which refers to any symptom profile similar to that of PD but with a known etiology such as vascular difficulties, Wilson’s disease, exposure to dopaminergic neurotoxins (e.g., MPTP), or encephalitis. The essential neuropathological changes in PD are a loss of melanine-containing dopaminergic neurons in the substantia nigra pars compacta (SNc) (cf. Hornykiewicz, 1998). This results in a dysfunction of the basal ganglia circuitry, which is an integral part of cortico-basal ganglia-cortical loops mediating motoric and cognitive functions. These pathophysiological changes, however, do not become clinically noticeable until a 60–80% level of striatal DA loss is achieved (Bernheimer, Birkmeyer, Hornykiewicz, Jellinger, &
B. Harel et al. / Brain and Cognition 56 (2004) 24–29
Seitenberger, 1973). Imaging studies have found an annual rate of reduction in striatal uptake of dopamine (via the dopamine transporter, or DAT) of about 6–13% in patients with PD versus only a 0–2.5% change in agematched control subjects (Marek et al., 2002). Hence, it is quite possible that PD subjects may exhibit sub-clinical prodromal features for a number of years prior to the onset of symptoms. If one assumes a linear trend (which has yet to be determined with any confidence), patients with a high loss of DAT—estimated to be as high as 13% per year—still might not become aware of increasing symptom severity until approximately 5 years following the onset of the hallmark neuropathological changes that are associated with PD. During this prodromal period these patients may exhibit subtle behavioral changes that might nonetheless be detectable with sensitive measurement techniques. There is currently no biomarker capable of definitively diagnosing PD, with the result being frequent underdiagnosis or misdiagnosis (Lang & Lozano, 1998). Autopsy studies have found inaccuracies in diagnosis in approximately one-quarter of the cases examined (Hughes, Daniel, Blankson, & Lees, 1993; Rajput, Rozdilsky, & Rajput, 1991). This is due to the fact that, as noted above, there are other diseases or states that manifest in clinically similar ways. The combination of asymmetric onset of symptoms, the presence of a 4–6 Hz resting tremor and/or a good response to the dopamine precursor, levodopa, is indicative of PD and can be used to differentiate PD from clinically similar diseases (e.g., multiple-system atrophy). However, some of these diseases present the same clinical signs and not all patients with PD show all of the ‘‘hallmark’’ signs of the disease. The ability to accurately distinguish PD from other diseases is important diagnostically and prognostically, especially when considering the development or application of new treatment therapies (e.g., new-generation neuroprotective agents and surgical techniques). One approach that has been considered by a number of investigators, for tracking disease severity with sufficient sensitivity, is the quantitative analysis of speech motor changes that have been associated with the disease (Darley, Brown, & Swenson, 1975; Mlcoch, 1992). Hypokinetic dysarthria, a speech disorder characterized by indistinctness of articulation, weakness of voice, lack of inflection, burst of speech, and hesitations and stoppages, is an integral part of the motoric changes in PD (Darley et al., 1975). It is generally believed that hypokinetic dysarthria occurs in at least half of all patients with PD (Mlcoch, 1992). In speech pathology clinics, 98% of all cases of hypokinetic dysarthria are the result of PD (Berry, 1983). Thus, hypokinetic dysarthria occurs often enough to serve as confirmatory evidence for the neurologic diagnosis and is sometimes the first sign of the disease (Duffy, 1995). The reason for this close association between voice changes and the onset of
25
PD is probably due to the fact that speech is arguably one of the most complicated motor actions under volitional control, and may therefore be susceptible to slight degenerative changes in basal ganglia circuitry affected in the pathophysiology of PD. Hypokinetic dysarthria refers to a particular pattern of speech motor dysfunction that results directly from the neuropathological effects of PD on the speech motor system. The relationship between speech motor changes and PD is reflected by changes in the acoustic measures of speech, which are the endpoints of a series of physiologic and anatomic changes that are initially caused by a loss of dopaminergic cells in the SNc. The loss of dopaminergic input into the striatum and the consequent dysregulation of the basal ganglia results in the cardinal motor deficits (resting tremor, muscular rigidity, bradykinesia, akinesia, and postural abnormalities) commonly seen in PD patients. These deficits adversely affect the three major anatomic subsystems (listed above) that govern speech motor control. Ultimately, such dysregulation manifests as a change in speech production that can be measured acoustically. For example, dopaminergic depletion in the SNc (the predominant neuropathological change in PD) commonly causes muscle rigidity (a cardinal motor deficit), which alters muscular control of the larynx (phonatory subsystem). This may lead to increased laryngeal tension (physiological correlate), which may lead to decreased fundamental frequency range and variability (F0 : lowest frequency of a periodic signal) of speech (Duffy, 1995). Acoustic parameters of speech in PD may be a sensitive measure of early disease progression due to the complexity inherent in the coordination of the speech motor subsystems that govern respiration, phonation, and articulation. Although there is a paucity of research relating changes in acoustic measures to disease progression in PD, evidence suggests that this may be the case (King, Ramig, Lemke, & Horii, 1994). Almost all studies of this association between speech and disease severity have relied on cross-sectional datasets (e.g., Homes, Oates, Phyland, & Hughes, 2000). For example, Canter (1963, 1965) noted a decrease in F0 range during syllable production and paragraph reading in 17 PD males off medication. Other investigators have also found a decrease in F0 range and variability during reading tasks (Flint, Black, Campbell-Taylor, Gailey, & Levinton, 1992; Metter & Hanson, 1986). Moreover, Metter and Hanson (1986) have shown that there appears to be an inverse relationship between F0 variability and both dysarthria severity and clinical disability. Based on prior research in this area and the complexity of the speech motor system, it is not unreasonable to suppose that it may be possible to detect an early change in one’s voice in the years leading up to the onset of symptoms. If so, then as new neuroprotective medicines become available in the coming years, these motor
26
B. Harel et al. / Brain and Cognition 56 (2004) 24–29
speech acoustical measures might offer a means to detect early treatment response before the point at which bothersome symptoms are manifested. With this pilot study, we sought to track changes in F0 variability over the prodromal period of PD (as well as for three years following formal diagnosis), in order to explore whether any such change in speech motor control is evident in the years prior to diagnosis. We chose to explore this question by obtaining speech samples that had been archived over an 11-year period, for a single individual with PD. These samples were compared to an age, sex, and occupation matched control subject for whom we also were able to obtain longitudinal speech samples. We hypothesized that F0 variability would decrease over the prodromal and initial phases of the illness (prior to any symptomatic treatment). This retrospective single-case study design was felt to be practical, at least as an initial attempt to identify possible speech changes during a lengthy prodromal phase of the illness, due to the low incidence of PD, the need to follow potential patients for approximately 10 years, and the substantial data processing requirements for data analyses.
2. Methods 2.1. Subjects An adult individual with PD was chosen based on the following criteria: (1) the existence of voice recordings from between 5 and 10 years prior to the diagnosis of PD to at least 2–3 years after the diagnosis; (2) these data were maintained in public archives and thus available to the public at large and considered to be within the public domain; (3) biographical information regarding the approximate date of diagnosis was available; and (4) the subject is a native speaker of American English. The control individual was without known neurological or psychological impairments and was matched to the PD individual with respect to sex, age, and profession. This study was approved by the Institutional Review Board of the University of Connecticut, and in order to protect the anonymity of the subjects, no identifying information, including dates of diagnosis, are reported. 2.2. Apparatus The voice recordings were digitized from VHS tapes supplied by CNN using ProTV 2.6.1 software and a Panasonic 4 Head HiFi Stereo Omnivision VHS VCR (PV-V4521), a Formac Studio TV/Video Digitizer (FS 1042-Oc), an Apple G4 Titanium desktop, and an Interlink 400mKII Audio Interconnect Cable. The data were then analyzed using an IBM T23 ThinkPad and
Kay Multi-speech Model 4400 software (Kay Elemetrics Corp., 2001).
3. Procedure 3.1. Data collection The video recordings were obtained from the archives of a leading national television news service on standard VHS tapes. We specifically requested a full listing of video recordings (a videography) of all edited news stories containing a formal interview or press conference with the individual of interest, which generally included a reporter’s narration and a SU (stand-up) at the conclusion. We selected two recordings per year from 10 years pre- to 5 years post-diagnosis for the individual with PD when available. Using the same time span for the matched control, two recordings were selected from 10 years pre-diagnosis, time of diagnosis, and 5 years post-diagnosis; this varied based on the availability of the recordings for a given subject. All recordings were selected based on their length; our intent was to maximize the amount of available speech recorded from the selected subject. This resulted in a total of 41 min and 22 s of recording for the individual with PD and 15 min and 18 s of recording for the control individual. These recordings were pre-filtered by CNN to remove background noise and 60 Hz machine noise. The VHS tapes were recorded digitally in QuickTime Player using ProTV 2.6.1 software. The sound track was then extracted from the digital file using QuickTime Player and exported to Sound Studio Classic as a 16 bit, 32 kHz, stereo WAV file. This file was then separated into smaller files each of which contained the speech recordings for one time period (month-day-year). Continuous speech clips of the subject speaking without interruption were then extracted from these files (for a listing of the number and length of speech clips by year see Table 1A). 3.2. Data analysis The files were analyzed using the Kay Multi-speech Model 4400 software package (Kay Elemetrics Corp., 2001), which permits editing and storage of acoustic samples. Each clip was displayed as an amplitude-bytime waveform. A cursor was placed at the beginning of the first word and a second cursor was placed at the end of the last word in that section of continuous speech. The trim selection option was then used to eliminate information outside of the section of interest. A subset of these speech clips were selected (see Table 1B) and acoustic measurements were made related to F0 variability during free speech. Speech clips were only included in the analysis if they met the following criteria:
B. Harel et al. / Brain and Cognition 56 (2004) 24–29
27
Table 1A Videotape data initially provided by the CNN Archives (number of individual video clips, and duration in seconds of recorded voice per year) Subject
Years from diagnosis of PD subject )7
)6
PD Control
)5
9 clips 1m 44s 4 clips 55s
)4
)3
)2
)1
Diagnosis
1
2
3
2 clips 28s
11 clips 2m 01s
7 clips 2m 02s
3 clips 59s
4 clips 1m 04s
4 clips 47s
15 clips 2m 14s
1 clip 09s
7 clips 1m 06s
4 clips 1m 30s
5 clips 30s
5 clips 1m 15s
Table 1B Videotape data used in analyses and represented in Fig. 1, following our quality control assessment (see Section 2) Subject
Years from diagnosis of PD subject )7
PD Control
)6 1 clip
)5
)4
)3
2 clips 3 clips
4 clips
(1) the samples contained emotionally neutral speech, that is, declarative as opposed to exclamatory or interrogative statements; (2) the speech clips were between 5 and 30 s in length (if a clip was over 30 seconds it was trimmed in order to standardize the length of the clips used across time points); and (3) the recordings were of suitable acoustic quality. Two methods were used in order to determine that the speech clips were of sufficient quality. The first consisted of perceptually evaluating each clip to ensure that there was no noticeable background noise (e.g., people speaking other than the interviewee). The second was done after the pitch contour was generated and consisted of excluding speech clips if F0 was calculated during pauses or unvoiced segments in speech, thus suggesting F0 calculation of recorded sound energy not solely related to speech. Each clip was displayed as an amplitude-by-time waveform. A wideband spectrogram was then generated and the Pitch Contour feature was used to calculate the F0 with a 25-ms analysis window for all portions of the
)2
)1
Diagnosis
1
2
3
1 clip 5 clips
2 clips
2 clips
3 clips 4 clips
1 clip 4 clips
voiced signal. The F0 variability, which tracks the degree of pitch variability within the selection, was derived mathematically by dividing the standard deviation by the mean. F0 variability was selected because dividing the standard deviation of F0 by the mean of F0 for a given individual creates a measure of percent change and thus can be used to compare between individuals or groups. A monotonic vocalization has lower variability.
4. Results The trajectory of fundamental frequency (F0 ) variability in free speech was constructed by plotting the mean F0 variability for each time point (i.e., year). As can be seen in Fig. 1, there is a decrease in F0 variability in the speech of this PD subject, starting several years prior to diagnosis and extending until the start of any pharmacologic intervention, which was not initiated until two years post-diagnosis. Conversely, F0 variability
Fig. 1. Fundamental frequency (F0 ) variability for PD subject and matched control during free speech.
28
B. Harel et al. / Brain and Cognition 56 (2004) 24–29
appears to be relatively stable in the control subject across time points. In addition, F0 variability seems to be comparable in the PD subject and the control subject prior to four years pre-diagnosis and after treatment began for the PD subject (two years post-diagnosis). To further explore these findings, the F0 variability data for each subject were divided into two groups based on years prior to diagnosis and following the initiation of symptomatic drug treatment for this specific PD subject: (1) the five year period leading up to diagnosis, and including the one year following diagnosis but before any drug treatment was initiated; and (2) samples obtained prior to the apparent five year prodrome (based on modeling of the rate of striatal DAT loss prior to diagnosis, as discussed above) and the two years after symptomatic drug treatment had been started. To assess the effects of time of speech data collection relative to the point at which the diagnosis was made, the measures of F0 variability for each separate speech sample, in each group of years, were compared in an Independent-Samples T Test for each subject. This analysis approach was chosen because each sample can be considered independent within the distribution of samples for each subject, based on the null hypothesis that any given F0 variability score should be equivalent to any score within a subject over the 11-year period. The F0 variability across the ‘group 1’ time points, compared to data obtained across the ‘group 2’ time points, was significantly lower in the PD subject (tð14Þ ¼ 7:88, p < :001, e ¼ :816). Conversely, no differences were found between the two sets of time points for the control subject (tð14Þ ¼ :63, p ¼ :540, e ¼ :027). To assess whether F0 variability is comparable between the two subjects, the measures of F0 variability for each separate speech sample, for each subject, were compared in an Independent-Samples T Test. Within the ‘group 1’ time points, the PD subject showed a significant decrease in F0 variability compared to the control subject (tð17Þ ¼ 5:54, p < :001, e ¼ :643). Conversely, within the ‘group 2’ time points, the PD subject showed a significant increase in F0 variability compared to the control subject (tð11Þ ¼ 2:31, p ¼ :041, e ¼ :327).
5. Discussion The principal goal of this single case-control study was to explore changes in F0 variability over the early course of PD, beginning prior to the onset of any readily observable clinical symptoms. For one PD subject, for whom we were able to obtain voice recordings extending over an 11-year period, we found that F0 variability over the entire 5 year period immediately preceding diagnosis and including one year post-diagnosis but prior to any symptomatic treatment (‘group 1’) was diminished relative to this individual’s initial baseline recording and
the recordings obtained after aggressive symptomatic intervention was initiated (‘group 2’). By comparison, the age, sex, and profession-matched healthy control subject showed no such changes in F0 variability between these same sets of time points. We also found that F0 variability, throughout the 5-year prodromal period and extending through the first year following diagnosis (and prior to any symptomatic treatment), was significantly lower in the PD subject compared to this same set of measurements for the control subject. Once symptomatic treatment was initiated for the PD subject (during the 2nd and 3rd years post-diagnosis), this individual’s measure of F0 variability appeared to ‘‘normalize’’ and return to a level that was more similar to that measured at six years prior to diagnosis (see Fig. 1). This pattern of results suggest that: (1) there is a decrease in free speech F0 variability that may be observed over approximately five years prior to clinical diagnosis, and we hypothesize that this change is related to a decrease in striatal dopamine transporter (DAT) population (which results from a progressive loss of dopaminergic neurons in the SNc); and (2) initiating symptomatic pharmacological treatment with a dopamine precursor (L -DOPA) serves to ameliorate this speech change with a normalization of F0 variability to a level that is comparable to that observed prior to a potential five year prodromal period. These tentative conclusions are based on both qualitative analysis of the data (Fig. 1), the exploratory statistical analyses described above, and a review of published literature suggesting that there may be a subclinical prodrome during which there is a steady drop in the DAT population (Bernheimer et al., 1973; Marek et al., 2002). There is also a considerable body of crosssectional research demonstrating a link between F0 variability in speech and the severity of symptoms in PD. For instance, PD patients have been shown to exhibit decreased F0 range and variability during reading tasks (Flint et al., 1992; Metter & Hanson, 1986) that further diminish as the severity of the disease increases (Metter & Hanson, 1986). A decrease in F0 range during paragraph reading has also been demonstrated in PD patients when L -DOPA treatment is temporarily withheld (Canter, 1963, 1965). Finally, preliminary findings from our lab indicate a significant decrease in F0 variability during free speech in three newly diagnosed and unmedicated PD patients compared to four age- and education-matched control subjects (tð5Þ ¼ 2:926, p ¼ :033). The results from this single case-control study should clearly be interpreted with caution, given the exploratory nature of this study and the consequent methodological limitations. These limitations include the use of only a single PD subject and a single matched control subject, an uncontrolled environment in which the speech data were collected, lack of a prospectively
B. Harel et al. / Brain and Cognition 56 (2004) 24–29
designed assessment protocol so that consistent speech samples can be collected and compared over time, and either a small amount of (or missing) data at certain time points. The results are also specific to this PD patient and may not be generalized to all PD patients. Nonetheless, given the obvious difficulties in designing and running a prospective study that would have to extend over at least one or more decades, we are encouraged by the possibilities afforded by this particular retrospective study design. To address some of these methodological limitations, we are currently in the process of collecting similar sets of video/audio data for additional PD and matched control subjects. Despite these methodological considerations, we believe that these results are promising and suggest that the quantitative analyses of voice acoustics may lead to a robust set of biomarkers for tracking both disease progression and treatment response in PD. References Bernheimer, H., Birkmeyer, W., Hornykiewicz, O., Jellinger, K., & Seitenberger, F. (1973). Brain dopamine and the syndromes of Parkinson and Huntington. Journal of Neurological Sciences, 20, 425–455. Berry, W. R. (1983). Clinical dysarthria. San Diego: College-Hill. Blandini, F., Giuseppe, N., Tassorelli, C., & Martignoni, E. (2000). Functional changes of the basal ganglia circuitry in Parkinson’s disease. Progress in Neurobiology, 62, 63–88. Canter, G. (1963). Speech characteristics of patients with Parkinson’s disease: I. Intensity, pitch, and duration. Journal of Speech and Hearing Disorders, 28, 221–229. Canter, G. (1965). Speech characteristics of patients with Parkinson’s disease: II. Physiological support for speech. Journal of Speech and Hearing Disorders, 30, 44–49. Darley, F. L., Brown, J. R., & Swenson, W. M. (1975). Language changes after neurosurgery for parkinsonism. Brain and Language, 2, 65–69.
29
Duffy, J. R. (1995). Motors speech disorders: substrates, differential diagnosis, and management. New York: Mosby. Flint, A., Black, S., Campbell-Taylor, I., Gailey, G., & Levinton, C. (1992). Acoustic analysis in the differentiation between Parkinson’s Disease and major depression. Journal of Psycholinguistic Research, 21, 383–399. Hoehn, M. M., & Yahr, M. D. (1967). Parkinsonism: Onset, progression, and mortality. Neurology, 17, 427–442. Homes, R. J., Oates, J. M., Phyland, D. J., & Hughes, A. J. (2000). Voice characteristics in the progression of Parkinson’s disease. International Journal of Language and Communication Disorders, 35, 407–418. Hornykiewicz, O. (1998). Biochemical aspects of Parkinson’s disease. Neurology, 51(Suppl. 2), S2–S9. Hughes, A. J., Daniel, S. E., Blankson, S., & Lees, A. J. (1993). A clinicopathologic study of 100 cases of Parkinson’s disease. Archives of Neurology, 50, 140–148. Kay Elemetrics Corp. (2001). Multi-Speech, Model 3700 CSL, Models 4100, 4300B, 4400 (v. 2.4). Lincoln Park, NJ: Kay Elemetrics Corp. King, J. B., Ramig, L. O., Lemke, J. H., & Horii, Y. (1994). Parkinson’s disease: Longitudinal changes in acoustic parameters of phonation. Journal of Medical Speech-Language Pathology, 2, 29–42. Lang, A. E., & Lozano, A. M. (1998). Parkinson’s disease. The New England Journal of Medicine, 339, 1044–1053. Marek, K., Seibyl, J, Shoulson, I, Holloway, R., Kieburtz, K., McDermott, M., Kamp, C., Shinaman, A., Fahn, S., Lang, A., Weiner, W., & Welsh, M. (2002). Dopamine transporter brain imaging to assess the effects of pramipexole vs levodopa on Parkinson disease progression. Journal of the American Medical Association, 287, 1653–1661. Metter, J., & Hanson, W. (1986). Clinical and acoustical variability in hypokinetic dysarthria. Journal of Communication Disorders, 19, 347–366. Mlcoch, A. G. (1992). Diagnosis and treatment of parkinsonian dysarthria. In W. C. Koller (Ed.), Handbook of Parkinson’s disease. New York: Marcel Decker. Parkinson, J. (1817). Essay on the shaking palsy. London: Sherwood, Neely, and Jones. Rajput, A. H., Rozdilsky, B., & Rajput, A. (1991). Accuracy of clinical diagnosis in parkinsonism—a prospective study. Canadian Journal of Neurological Sciences, 18(3), 275–278.
