Uniform data set language measures for bvFTD and PPA diagnosis and monitoring.

INTRODUCTION: The Frontotemporal Lobar Degeneration Module (FTLD-MOD) includes a neuropsychological battery designed to assess the clinical features of FTLD, although much is unknown about its utility. We investigated FTLD-MOD and Uniform Data Set 3.0 (UDS) language tests for differential diagnosis and disease monitoring.
METHODS: Linear regressions compared baseline performances in 1655 National Alzheimer's Coordinating Center participants (behavioral variant frontotemporal dementia (bvFTD, n = 612), semantic variant primary progressive aphasia (svPPA, n = 168), non-fluent/agrammatic variant PPA (nfvPPA, n = 168), logopenic variant PPA (lvPPA, n = 109), and controls (n = 581)). Sample sizes to detect treatment effects were estimated using longitudinal data.
RESULTS: Among PPAs, the FTLD-MOD language tasks and UDS Multilingual Naming Test accurately discriminated svPPA. Number Span Forward best discriminated lvPPA; Phonemic:Semantic Fluency ratio was excellent for nfvPPA classification. UDS fluency and naming measures required the smallest sample size to detect meaningful change. DISCUSSION: The FTLD-MOD and UDS differentiated among PPA subtypes. UDS 3.0 measures performed best for longitudinal monitoring.

BACKGROUND

Frontotemporal dementia (FTD) is a collection of clinical syndromes that present with cognitive and motor impairments. Diverse FTD syndromes often arise from a non‐Alzheimer's pathologic substrate, known as frontotemporal‐lobar degeneration (FTLD), with most cases caused by FTLD‐tau or FTLD with inclusions oftransactive response DNA‐binding protein (FTLD‐TDP). However, the clinical syndromes often do not map consistently to specific pathologic entities. Behavioral variant FTD (bvFTD) is typified by early loss of social decorum and behavioral changes, with relative preservation of language, although language impairments can be present. Of the FTD clinical syndromes, pathological prediction is the most challenging in bvFTD. The FTD spectrum includes two language syndromes, classified as variants of primary progressive aphasia (PPA). Patients with the semantic variant (svPPA) exhibit a loss of semantic knowledge with relative preservation of speech production, grammar, and repetition. Most svPPA cases are associated with FTLD‐TDP. The primary clinical features of the non‐fluent/agrammatic variant (nfvPPA) are effortful spontaneous speech, agrammatism, and errors in speech production with spared semantic knowledge; an nfvPPA clinical syndrome is most predictive of FTLD‐tau. , A third PPA syndrome, logopenic variant PPA (lvPPA), presents with impaired repetition and single‐word retrieval, with relative preservation of semantic knowledge and grammar. , Unlike the other PPA variants, lvPPA has been associated with Alzheimer's disease (AD) pathology in 80% to 90% of cases, , although other studies have reported FTLD pathology in 44% to 60%. , Thus correct syndromic classification can serve as a guide to predicting underlying pathology, albeit with probabilistic accuracy.

Differentiating bvFTD and the PPA subtypes can present a substantial diagnostic challenge. Although as common as AD in early onset neurodegenerative disease, FTD is still relatively rare and provides fewer opportunities for clinicians to sharpen their diagnostic acumen. Nevertheless, accurate diagnosis is important for clinical care and treatment planning. The value of differential diagnosis will continue to grow as therapeutic treatments become available. Unlike AD, there are currently no molecular biomarkers to differentiate FTLD‐tau and FTLD‐TDP; clinical syndromes, although imperfect, remain one of the best in vivo predictors of pathology.

The differential diagnosis of FTD and PPA requires not only a detailed clinical history, but also a comprehensive evaluation of speech, language, cognition, and behavior. Many FTD and PPA patients in the United States have been studied through the Alzheimer's Disease Centers (ADC) program of the National Institute on Aging (NIA). These participants complete a battery of neuropsychological measures as part of the Uniform Data Set (UDS). A study of pathologically verified FTLD and AD patients, however, concluded that an earlier version of UDS did not adequately differentiate between FTLD and AD. In response, a battery of tests called the FTLD Module (FTLD‐MOD) was designed to capture the core features of bvFTD and PPA. This battery comprises several measures of language, social cognition, and behavior. An initial investigation of biomarker‐supported bvFTD and AD suggested that the FTLD‐MOD improves discrimination of these groups. In a recent study, clinically diagnosed PPA participants performed worse than bvFTD on all FTLD‐MOD language tasks. The utility of these tasks for discriminating among the various PPA subtypes, however, remains unknown. In addition, the newest version of the UDS (v3.0; henceforth, UDS) includes several language tests relevant to FTLD.

In addition to validating tests for differential diagnosis, there is an urgent need for measures that are sensitive to changes over time in PPA and bvFTD. These measures could serve as endpoints for treatment trials and help clinicians monitor cognition. Sensitive endpoints are especially important in rare conditions such as FTLD. Although several functional and cognitive endpoints have been identified in PPA and bvFTD, , there remains room for improvement. , FTLD‐MOD and UDS tests may be candidates for longitudinal tracking. The present study investigated the utility of the FTLD‐MOD and UDS language tests for differential diagnosis in PPA and longitudinal monitoring in bvFTD and PPA.

METHODS

Participants

The sample included participants from the National Alzheimer's Coordinating Center (NACC) database who completed at least one FTLD‐MOD language task, between February 2012 and February 2020, at NIA‐funded ADCs and affiliated centers. Beginning in 2014, many PPA and bvFTD participants were recruited into the Advancement of Research and Treatment in Frontotemporal Lobar Degeneration (ARTFL; U54 NS092089) +/‐ Longitudinal Evaluation of Familial Frontotemporal Dementia Subjects (LEFFTDS: U01 AG045390) programs and co‐enrolled into the respective site ADCs. Participants were excluded if their first language was not English. All patients were tested in English. All participants (or their proxies) provided written informed consent.

Controls were defined as those having normal cognition according to UDS protocol and a global Clinical Dementia Rating Scale plus NACC FTLD module (CDR®+NACC FTLD) , score of 0 at all visits. PPA or bvFTD syndromes were diagnosed according to research criteria , without considering FTLD‐MOD performance. Patients with co‐diagnosed bvFTD and PPA were diagnosed according to their PPA syndrome (svPPA = 79;nfvPPA = 21;lvPPA = 8). These criteria resulted in 1655 participants, including 612 participants with a primary clinical diagnosis of bvFTD, 185 with svPPA, 168 with nfvPPA, 109 with lvPPA, and 581 cognitively normal controls. The same pattern of results reported in this manuscript was observed after restricting the sample to cases without co‐diagnoses.

Measures

CDR®+NACC‐FTLD

The CDR®+NACC‐FTLD is scored similarly to the traditional CDR®: a total score is calculated to categorize each patient as having mild features of neurodegenerative disease (CDR®+NACC‐FTLD = 0.5) or clear features of an overt neurodegenerative syndrome (CDR®+NACC‐FTLD = 1, 2, or 3). The CDR®+NACC‐FTLD includes two additional domains germane to FTLD: Behavior and Language. , The eight domain scores were summed to calculate the Sum of Boxes (range:0‐24; higher scores indicate greater impairment). ,

RESEARCH IN CONTEXT

Systematic Review: The authors reviewed the existing literature on the Frontotemporal Lobar Degeneration Module (FTLD‐MOD) and Uniform Data Set (UDS), version 3.0. Prior studies investigated the FTLD‐MOD in cross‐sectional samples. A recent study found the FTLD‐MOD differentiated between behavioral variant frontotemporal dementia (bvFTD) and primary progressive aphasia (PPA). No prior studies assessed FTLD‐MOD performance across the three PPA subtypes, and none analyzed longitudinal FTLD‐MOD performance.

Interpretation: The FTLD‐MOD and UDS language tests distinguished PPA from bvFTD and discriminated the semantic variant and non‐fluent variants from other PPA groups with excellent accuracy. Discrimination was acceptable for the logopenic variant. Fluency and naming measures from the standard UDS 3.0 neuropsychological battery offered the best cross‐sectional discrimination of PPA and lowest sample size estimate needed to detect clinically relevant change all FTD subtypes.

Future Direction: More sensitive and specific measures would be useful for diagnosis and monitoring of the logopenic and non‐fluent variants in research studies.

FTLD‐MOD language measures

The FTLD‐MOD primarily comprises measures of language and social cognition/behavior. The social cognition measures are detailed in several published , , , and ongoing studies. For FTLD‐MOD documentation, visit: https://www.alz.washington.edu.

Regular and irregular word reading

Participants first read a list of 15 regularly spelled words (ie, adhering to standard phonic rules) aloud. Participants then read 15 irregularly spelled words (eg, gnome, yacht). Irregular word reading requires semantic knowledge or familiarity. Making “regularization” errors when reading these words, termed surface dyslexia, is featured in the svPPA diagnostic criteria. Regular word reading was hypothesized to be more impaired in lvPPA and nfvPPA secondary to impaired phonological processing (lvPPA) or speech production (nfvPPA).

Sentence repetition and reading

Participants first repeat five sentences that are presented orally. Participants are later provided with these five sentences in a written format and asked to read them aloud. Repetition impairment is a core feature of lvPPA and relatively preserved in svPPA. Repetition may be impaired in nfvPPA due to impairments in motor speech, apraxia of speech, grammatical processing errors, or executive dysfunction.

Northwestern Anagram Test (NAT)

The NAT requires participants to organize words to create 10 grammatically correct sentences that describe a picture stimulus, allowing for assessment of sentence production independent of speech production, word‐finding difficulties, or working memory capacity. This test was hypothesized to detect the grammatic impairments often seen in nfvPPA.

Semantic Word‐Picture Matching Test

Participants hear a word and then choose the picture that matches this word from one of four semantically related stimuli. This 20‐item task of single‐word comprehension requires lexical‐semantic knowledge, and similar tasks are consistently impaired in svPPA. ,

Semantic Association Test

Participants are presented with two pairs of pictures (eg, squirrel‐tree, squirrel balloon) and are asked to choose the pair with semantically related objects (Northwestern Naming Battery; NNB, 16 trials). We expected svPPA patients to perform poorly on this task, as they consistently show impairment on similar tasks , of semantic knowledge. , , , ,

Noun and verb naming

In this NNB naming subtest, participants are first presented with 16 drawings of objects (ie, nouns) followed by 16 drawings of action items (ie, verbs). The outcome is total correct. Noun naming was hypothesized to require temporal lobe–mediated semantic knowledge and, therefore, svPPA patients were expected to show the greatest degree of impairment. Naming verbs has frontal and posterior parietotemporal correlates and was anticipated to be impaired in nfvPPA based on prior findings.

Other UDS language measures

Phonemic and semantic fluency

Phonemic fluency requires participants to generate as many words as possible in 60 seconds that begin with two letters (“F” and “L”). The outcome is total correct, summed across both trials. In semantic/category fluency, participants have 60 seconds to produce words belonging to two different semantic categories (“animals” and “vegetables”). We calculated the ratio of Phonemic:Semantic Fluency, as svPPA participants display relatively greater difficulty with semantic fluency compared to phonemic fluency, , , whereas nfvPPA participants evidence the opposite pattern. ,

Multilingual Naming Test (MINT)

This 32‐item object picture naming task was designed to assess naming skills in speakers of multiple languages. Outcome was total score, including items named correctly with semantic, but not phonemic cues.

Number span forward

Participants are read increasingly longer sequences of numbers and asked to repeat them immediately in the order presented. This measure of phonological loop function was anticipated to help discriminate lvPPA.

Statistical analysis

Continuous demographic variables (ie, age, education) were compared between diagnostic groups using two separate linear models with a categorial predictor (ie, diagnostic group) and the demographic variable as the outcome, followed by pairwise group contrasts. A chi‐square test was used to assess group differences in sex.

A cross‐sectional comparison of FTLD‐MOD performance among controls, bvFTD, and PPA variants was conducted using a multivariable linear model, with test performance as the outcome and diagnostic group as the categorical predictor of interest, adjusting for age, sex, education, and disease severity (Sum of Boxes). Predictors were chosen a priori and entered simultaneously. Post hoc group comparisons were made with Bonferroni multiple comparison correction. Disease severity was not a covariate in comparisons between patient groups and controls, as this would obscure the effect of interest.

In a follow‐up, cross‐sectional analysis, we estimated the potential use of these measures for differential diagnosis by fitting logistic regressions and receiver‐operating characteristic (ROC) curves. We first selected promising measures for each of the three PPA variants based on results of pairwise group comparisons (ie, those differentiating groups after Bonferroni correction, or in the case of nfvPPA, measures in which nfvPPA performed the worst). We estimated the area under the curve (AUC) for differentiating each PPA variant from the other two PPA groups. We also compared lvPPA and nfvPPA directly, given that this is often the most challenging clinical comparison due to overlapping clinical features.

Longitudinal rates of decline were estimated by fitting linear mixed‐effects models with random intercepts and slopes. Group differences in slopes were assessed by including a diagnosis‐by‐time interaction term with post hoc pairwise comparisons (Bonferroni corrected). Time was included as a continuous variable indicating years from baseline assessment. Models controlled for baseline age, sex, education, and severity (Sum of Boxes). For illustrative purposes, subject‐specific slopes were extracted and plotted. Statistical tests of between‐group differences were conducted using mixed‐effects models, not by performing statistics on the subject‐specific extracted slopes, as they are biased by shrinkage.

Sample size estimates to detect a 40% and 25% treatment effects (40% or 25% reduction in decline from timepoint 1 to 2, adjusted for 20% attrition) were derived based on the annualized change score between baseline and follow‐up. For each measure, participants were included if they had received a second assessment of that measure within 2.5 years of their baseline visit.

RESULTS

Baseline demographics and clinical characteristics are displayed in Table 1. Groups differed on age and sex, but not education. Post hoc comparisons (all P‐values Bonferroni‐adjusted) showed that all four patient groups were older than controls (P < .001). LvPPA and nfvPPA were older than bvFTD (P‐values < .05). NfvPPA were older than svPPA (P = .022), with no significant differences between svPPA and lvPPA (P = .196) or between lvPPA and nfvPPA (P = 1.0). Group differences were observed in CDR®+NACC FTLD Sum of Boxes, with svPPA and bvFTD having the greatest box scores.

TABLE 1

Sample description, baseline demographics, and clinical characteristics: M (SD)

	Controls	bvFTD	svPPA	nfvPPA	lvPPA	P‐value
N	581	612	185	168	109
N with follow‐up visits	224	233	79	65	47
Number of follow‐up visits	2.9 (1.2)	2.9 (1.4)	2.7 (0.9)	2.7 (0.8)	2.7 (1.0)
Length of follow‐up (y)	2.6 (1.6)	2.3 (1.7)	2.1 (1.2)	1.8 (0.9)	1.9 (1.1)
Age (y)	50.2 (15.8)	63.7 (8.5)	65.0 (7.6)	67.9 (8.2)	67.8 (8.2)	<.001
Education (y)	17.2 (10.6)	17.3 (11.4)	17.4 (10.8)	17.7 (13.0)	17.0 (8.3)	0.27
Sex (% male)	42.9	62.4	48.7	47	54.13	<.001
CDR® + NACC FTLD
Sum of Boxes	0.0 (0.0)	8.4 (4.2)	7.6 (4.7)	5.1 (4.0)	5.1 (3.5)	<.001
Global Score (N, %)
0	581 (100)	–	–	–	–
0.5	–	39 (6.4)	35 (18.9)	59 (35.1)	37 (33.9)
1	–	139 (22.7)	52 (28.1)	43 (25.6)	38 (34.9)
2	–	367 (60.0)	80 (43.2)	56 (33.3)	29 (26.6)
3	–	67 (11.0)	18 (9.7)	10 (6.0)	5 (4.6)

Note. Length of follow‐up and number of follow‐up visits calculated for participants with more than one time point.

Abbreviations: bvFTD = behavioral variant frontotemporal dementia; svPPA = semantic variant primary progressive aphasia; nfvPPA = nonfluent variant PPA; lvPPA = logopenic variant PPA; y = years; CDR® + NACC FTLD = Clinical Dementia Rating Scale plus National Alzheimer's Coordinating Center FTLD Module

Baseline performances

Ceiling effects were noted for controls on all measures (Table 2; Supplemental Figure S1). After Bonferroni correction, all diagnostic groups performed worse than controls on nearly every FTLD‐MOD measure, with the exception of Word‐Picture Matching. Similarly, on the majority of measures, all PPA groups performed worse than bvFTD, with a few notable exceptions. On several measures (eg, Noun Naming, Word‐Picture Matching, Semantic Associations) only the svPPA group performed consistently worse than bvFTD. Only the lvPPA group performed worse than bvFTD on the NAT.

TABLE 2

Baseline FTLD‐MOD subtest raw scores by diagnostic group

Instrument	Controls	bvFTD	svPPA	nfvPPA	lvPPA	Lower than controls	Lower than bvFTD	PPA comparison
	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Word reading – regular	15.0 (0.1)	14.7 (0.9)	13.9 (2.4)	13.8 (3.1)	14.2 (2.4)	All	sv, nfv, lv	sv = nfv = lv
Word reading – irregular	14.5 (0.8)	13.2 (2.4)	8.9 (3.9)	12.0 (3.6)	11.5 (3.2)	All	sv, nfv, lv	sv < (nfv = lv)
Sentence repetition	4.5 (0.6)	4.0 (1.1)	3.5 (1.4)	2.9 (1.7)	2.2 (1.7)	All	sv, nfv, lv	lv < nfv < sv
Sentence reading	4.9 (0.4)	4.4 (1.0)	3.9 (1.3)	3.4 (1.7)	3.9 (1.4)	All	sv, nfv, lv	nfv < (sv = lv)
NW anagram total	9.3 (1.4)	6.9 (2.7)	6.9 (2.9)	6.5 (3.0)	5.8 (2.5)	All	nfv, lv	(nfv = lv) < sv
Word‐picture matching	20.0 (0.1)	19.3 (1.8)	17.2 (3.2)	18.9 (3.1)	19.4 (1.3)	bv, sv, nfv	sv, nfv, lv	sv < (nfv = lv)
Semantic associations	16.0 (0.2)	14.4 (2.7)	12.7 (3.0)	14.9 (2.6)	15.0 (2.3)	All	sv	sv < (nfv = lv)
Noun naming	15.9 (0.3)	14.8 (2.3)	8.5 (5.3)	14.7 (2.8)	13.1 (3.6)	All	sv, nfv, lv	sv < lv < nfv
Verb naming	15.9 (0.3)	14.2 (2.8)	10.5 (4.5)	13.3 (4.3)	11.7 (4.3)	All	sv, nfv, lv	(sv = lv) < nfv
Other UDS measures
Phonemic fluency	29.8 (7.7)	15.9 (9.9)	14.1 (7.4)	11.0 (7.1)	14.5 (6.6)	All	nfv, lv	(sv = lv), (nfv = lv), nfv < sv
Semantic fluency	38.9 (7.9)	19.3 (10.6)	10.7 (7.4)	18.3 (9.9)	13.5 (7.6)	All	sv, nfv, lv	(sv = lv) < nfv
Phonemic: Semantic ratio	0.8 (0.2)	0.9 (0.6)	2.0 (2.6)	0.6 (0.3)	1.2 (0.9)	None a	None a	nfv < lv < sv
MINT	30.3 (1.7)	24.9 (6.9)	8.6 (8.7)	26.5 (6.7)	19.9 (8.9)	All	sv, lv	sv < lv < nfv
Number span forward	9.0 (2.3)	6.7 (2.4)	6.5 (2.7)	5.0 (2.3)	3.8 (2.2)	All	nfv, lv	(nfv = lv) < sv

Note. Group comparisons include Bonferroni adjustments for multiple comparisons; P < .05.

All = all four diagnostic groups.

svPPA was significantly higher than controls and bvFTD on Phonemic:Semantic ratio.

Abbreviations: bvFTD = behavioral variant frontotemporal dementia; svPPA = semantic variant primary progressive aphasia; nfvPPA = nonfluent variant PPA; lvPPA = logopenic variant PPA; NW = Northwestern; UDS = Uniform Data Set; MINT: Multilingual Naming Test

Several tests of semantic knowledge consistently differentiated svPPA from the other diagnostic groups. ROC analyses testing the discrimination of svPPA from the other two PPA variants suggested excellent accuracy for Noun Naming (area under the curve [AUC] = 0.82 [0.78, 0.86]), and acceptable discrimination for Semantic Associations (AUC = 0.77 [0.73, 0.82]), Irregular Word Reading (AUC = 0.74 [0.69, 0.78]), and Word‐Picture Matching (AUC = 0.73 [0.68, 0.78]). When all four predictors were added to the same logistic regression, Noun Naming (P < .001) and Semantic Association (P = .043) remained as independent predictors of diagnosis. Including both tasks in the ROC analysis, however, did not provide a clinically significant improvement in accuracy (AUC = 0.83 [0.79, 0.88]) compared to Noun Naming alone. The MINT (Figures 1A and 2A) outperformed all FTLD‐MOD tasks in discriminating svPPA (AUC = 0.89 [0.85, 0.93]), although there was overlap in the 95% confidence intervals.

FIGURE 1

Uniform Data Set language measures for differential diagnosis and longitudinal monitoring of frontotemporal dementia (FTD) syndromes. Note. Figures 1A and B present the most promising cross‐sectional (A) and longitudinal (B) measures for semantic variant primary progressive aphasia (PPA). Figures 1C and D present the most promising cross‐sectional (C) and longitudinal (D) measures for non‐fluent/agrammatic variant PPA. Figures 1E and F present the most promising cross‐sectional (E) and longitudinal (F) measures for logopenic variant PPA. The most promising cross‐sectional measures were determined based on receiver‐operating characteristic curves, shown in Figure 2. The most promising longitudinal measures were determined based on estimates of the sample sizes needed to detect a 40% treatment effect (Table 4). The y‐axis in 1C was truncated for illustrative purposes; four values were cut off from the svPPA group (10, 12, 14, and 19)

FIGURE 2

Classification accuracy for the primary progressive aphasia (PPA) variants. This figure displays tests with the highest accuracy for discriminating each of the three PPA variants. (A) The receiver‐operating characteristic (ROC) curve for the performance Multilingual Naming Test (MINT) in discriminating semantic variant (svPPA) from either non‐fluent agrammatic (nfvPPA) or logopenic variant (lvPPA). (2) The ROC curve for the performance of the ratio of Phonemic:Semantic Fluency in classifying nfvPPA compared to lvPPA and svPPA. Lower ratios are observed in nfvPPA. (C) The ROC curve for the performance of Number Span Forward in classifying lvPPA compared to nfvPPA and svPPA

TABLE 4

Estimated sample sizes for treatment trials

		bvFTD		svPPA		nfvPPA		lvPPA
Instrument	Effect Size (%)	n	Sample size	n	Sample size	n	Sample size	n	Sample size
Word reading ‐ regular	40	218	5540	78	1593	43	1883	45	2390
	25		14173		4070		4815		6115
Word reading ‐ irregular	40	217	2968	77	565	43	1125	45	3260
	25		7590		1443		2875		8340
Sentence repetition	40	215	2300	72	1818	41	1495	42	918
	25		5883		4650		3820		2345
Sentence reading	40	213	6385	72	1738	40	4753	42	5033
	25		16338		4445		12160		12878
NW anagram total	40	173	4248	54	3195	48	5298	34	22613
	25		10870		8170		13553		57880
Word‐picture matching	40	227	2943	75	405	55	5138	47	3050
	25		7528		1030		13148		7800
Semantic associations	40	209	1615	64	1305	53	4158	45	1650
	25		4133		3338		10640		4223
Noun naming	40	219	1875	66	260	44	5515	38	770
	25		4795		663		14110		1968
Verb naming	40	218	1790	66	310	44	8795	38	683
	25		4575		785		22508		1738
Other measures
CDR®+NACC FTLD SB	40	227	500	78	575	58	775	48	388
	25		1275		1465		1980		983
Phonemic fluency	40	153	898	44	2878	27	883	27	208
	25		2293		7360		2255		528
Semantic fluency	40	208	673	69	323	43	518	40	980
	25		1718		823		1320		2505
Phonemic: Semantic ratio	40	148	44498	35	1833	24	4053	27	3965613
	25		113908		4685		10370		10151965
MINT	40	156	1025	34	198	32	1195	27	448
	25		2620		500		3050		1138
Number span forward	40	156	1715	46	2240	29	840	29	2590
	25		4388		5730		2145		6628

Note. Estimates are for the total sample size required for a clinical trial (both arms) to detect “moderate” (40%) or “small” (25%) treatment effects, accounting for 20% expected attrition.

n is the number of participants that went into the estimate.

bvFTD, behavioral variant frontotemporal dementia; lvPPA, logopenic variant PPA; nfvPPA, non‐fluent variant PPA; PPA, primary progressive aphasia; svPPA, semantic variant PPA; Sent = Sentence; NW = Northwestern; CDR®+NACC FTLD SB = Clinical Dementia Rating Scale plus National Alzheimer's Coordinating Center FTLD Module Sum of Boxes; MINT: Multilingual Naming Test

The non‐fluent/agrammatic and logopenic variants can often be difficult to discriminate in clinical practice. Similarly, their performance on many measures were not statistically different at the group level. Sentence Reading appeared to be the most promising FTLD‐MOD measure for differentiating nfvPPA from the other two PPA syndromes. ROC analysis, however, suggested minimal utility for this task in discriminating nfvPPA from svPPA and lvPPA (AUC = 0.55 [0.49, 0.61]) and nfvPPA from lvPPA (AUC = 0.55 [0.48, 0.62]). NfvPPA participants were the worst performing group on Phonemic Fluency, although their performance did not differ significantly from lvPPA. A ratio of Phonemic:Semantic Fluency (Figures 1C and 2B) provided excellent accuracy in classifying nfvPPA from svPPA and lvPPA (AUC = 0.84 [0.79, 0.89]), and from lvPPA only (AUC = 0.82, [0.75, 0.90]).

The FTLD‐MOD test that performed best at discriminating lvPPA was Sentence Repetition, which was most impaired in lvPPA. ROC analysis revealed only minimal accuracy in separating lvPPA from the other two PPA syndromes (AUC = 0.67 [0.60, 0.72]) and from nfvPPA (AUC = 0.61 [0.54, 0.69]). Acceptable discrimination was observed for Number Span Forward in classifying lvPPA compared to svPPA and nfvPPA (Figures 1E and 2C; AUC = 0.72 [0.64, 0.79]), but only minimal accuracy for classifying lvPPA versus nfvPPA (AUC = 0.64 [0.55, 0.74]).

Longitudinal performances and sample size estimates

Annualized rates of decline for each diagnostic group are presented in Table 3, along with group comparisons. Similar to the cross‐sectional results, the most consistent differences in longitudinal trajectories were observed for the svPPA cases (Table 3; Figure 1C and D). To understand the effect sizes of longitudinal decline, we calculated sample sizes for a planned clinical trial (Table 4). Of the FTLD‐MOD tasks, the most encouraging results were observed for svPPA, with Noun and Verb Naming and Word‐Picture Matching showing the most promise. Indeed, these measures suggest a trial in svPPA would require only 260 participants to detect a 40% treatment effect if Noun Naming was used, compared to 575 if Sum of Boxes was used. Of note, for all diagnoses, standard UDS language measures outperform the FTLD‐MOD with regard to sample size (Table 4). Longitudinal trajectories for the best‐performing measures in each variant, based on sample size estimates, are displayed in Figure 1B, D, and F.

TABLE 3

Raw annualized change and adjusted group comparisons

	Controls	bvFTD	svPPA	nfvPPA	lvPPA
Instrument	Annual change (SE)	Annual change (SE)	Annual change (SE)	Annual change (SE)	Annual change (SE)	Worse than controls	Worse than bvFTD	PPA comparison
Word reading ‐ regular	0.0003 (0.00)	−0.05 (0.04)	−0.87 (0.20)	−0.50 (0.22)	−0.38 (0.16)	bv, sv	sv	(sv = lv), (lv = nfv), sv < nfv
Word reading ‐ irregular	0.019 (0.01)	−0.13 (0.04)	−1.32 (0.15)	−0.70 (0.20)	−1.16 (0.23)	all	lv, sv	(sv = lv), (lv = nfv), sv < nfv
Sentence repetition	−0.0003 (0.01)	−0.08 (0.03)	−0.39 (0.09)	−0.36 (0.13)	−0.55 (0.08)	lv, sv, nfv	lv, sv	sv = lv = nfv
Sentence reading	0.005 (0.01)	−0.07 (0.02)	−0.41 (0.09)	−0.30 (0.13)	−0.48 (0.12)	all	lv, sv	sv = lv = nfv
NW anagram total	0.04 (0.02)	0.00 (0.06)	−0.37 (0.15)	−0.61 (0.23)	−0.53 (0.16)	sv, nfv	sv, nfv	sv = lv = nfv
Word‐picture matching	0.001 (0.00)	−0.35 (0.08)	−1.37 (0.14)	−0.28 (0.16)	−0.56 (0.23)	bv, lv, sv	sv	(sv = lv), (lv = nfv), sv < nfv
Semantic associations	−0.003 (0.01)	−0.65 (0.11)	−0.86 (0.15)	−0.23 (0.10)	−0.53 (0.17)	bv, lv, sv	none	(sv = lv), (lv = nfv), sv < nfv
Noun naming	0.004 (0.00)	−0.68 (0.11)	−2.32 (0.22)	−0.31 (0.21)	−1.36 (0.30)	bv, lv, sv	sv	(sv = lv) < nfv
Verb naming	0.013 (0.01)	−1.01 (0.15)	−2.18 (0.23)	−0.80 (0.25)	−1.36 (0.26)	all	sv	(sv = lv), (lv = nfv), sv < nfv
Other UDS measures
Phonemic fluency	0.59 (0.16)	−1.18 (0.23)	−1.46 (0.36)	−1.89 (0.46)	−2.10 (0.40)	all	none	sv = lv = nfv
Semantic fluency	−0.13 (0.13)	−2.38 (0.25)	−3.29 (0.27)	−3.32 (0.40)	−2.49 (0.57)	all	none	sv = lv = nfv
Phonemic Semantic ratio	0.02 (0.005)	0.01 (0.01)	0.23 (0.15)	0.01 (0.02)	0.0001 (0.06)	sv a	sv a	(lv = nfv) < sv a
MINT	0.14 (0.04)	−1.51 (0.25)	−2.38 (0.46)	−1.28 (0.59)	−3.52 (0.68)	all	lv	(lv = nfv), (nfv = sv), (lv < sv)
Number span forward	0.03 (0.05)	−0.16 (0.06)	−0.59 (0.12)	−0.66 (0.19)	−0.60 (0.12)	all	sv, nfv	sv = lv = nfv

Note. Annualized changes scores were derived from linear mixed‐effects models.

Group comparisons among diagnostic groups control for baseline age, baseline CDR®+NACC FTLD Sum of Boxes, Education, and Sex.

Post hoc comparisons with controls include demographic covariates but do not control for disease severity (CDR® + NACC FTLD Sum of Boxes).

“Worse than controls” indicate more steeper longitudinal decline compared to controls.

“Worse than bvFTD” indicates a steeper longitudinal decline compared to bvFTD.

svPPA showed more rapid increases in the ratio of Phonemic:Semantic fluency compared to all other groups.

Abbreviations: SE = standard error; bvFTD = behavioral variant frontotemporal dementia; svPPA = semantic variant primary progressive aphasia; nfvPPA = nonfluent variant PPA; lvPPA = logopenic variant PPA; Sent = Sentence; NW = Northwestern; UDS = Uniform Data Set; MINT: Multilingual Naming Test

DISCUSSION

Clinical care and clinical trials in FTLD require measures that accurately differentiate FTD clinical syndromes and are sensitive to longitudinal change. The FTLD‐MOD has shown promise for differentiating between the clinical syndromes of bvFTD and PPA. Our study extended prior work by investigating the utility of FTLD‐MOD and UDS language tasks for cross‐sectional differential diagnosis among the PPA subtypes and their potential for longitudinal monitoring using a large publicly available data set. We confirmed prior findings that PPA participants generally performed worse than bvFTD, and the bvFTD group also performed statistically worse than controls on nearly every measure.

Cross‐sectional analyses indicated that a subset of FTLD‐MOD language tests consistently differentiate svPPA from controls, bvFTD, and other PPA subtypes. Noun Naming showed the most promise for differentiating svPPA from lvPPA and nfvPPA (AUC = 0.82). Word‐Picture Matching (AUC = 0.73) and Semantic Association (AUC = 0.77) were also promising. Another naming test from the standard UDS, the MINT, evidenced excellent discrimination of svPPA from other PPA syndromes (AUC = 0.89); adding additional FTLD‐MOD measures to the model did not strengthen prediction. The MINT was previously found to discriminate AD and amnestic mild cognitive impairment (MCI) from controls.

FTLD‐MOD measures showed less promise for discriminating lvPPA and nfvPPA. Although Sentence Repetition performance was worst in lvPPA, as expected, the ROC analysis suggested minimal diagnostic utility (AUC = 0.67). A similar finding was observed for nfvPPA: although Sentence Reading was most impaired in nfvPPA, it showed little promise as a diagnostic measure (AUC = 0.55). Patterns of performance on UDS language measures were consistent with the hypothesized directions, with Number Span Forward performing the best for classifying lvPPA (AUC = 0.72) and the ratio of Phonemic:Semantic Fluency for nfvPPA (AUC = 0.84); of interest, the ratio of Phonemic:Semantic Fluency also performed well at classifying nfvPPA versus lvPPA (AUC = 0.82).

The results of the longitudinal analysis paralleled the cross‐sectional analysis in several ways. Most groups showed greater longitudinal decline on all measures compared to controls, and in many cases, compared to bvFTD. SvPPA emerged as the group that was most accurately discriminated and had the lowest sample size estimates needed to detect a 40% treatment effect. Noun Naming and Word‐Picture Matching had reasonable sample size estimates (260 and 405, respectively). Consistent with the cross‐sectional analysis, longitudinal UDS measures such as the MINT (n = 198) outperformed the FTLD‐MOD measures in svPPA. We also replicated prior work showing promising sample sizes using Semantic Fluency for svPPA. Also analogous to the cross‐sectional results, the FTLD‐MOD tasks performed less well at longitudinally differentiating lvPPA and nfvPPA. Sample size estimates were unrealistic for all FTLD‐MOD tasks in lvPPA and nfvPPA, with UDS measures performing best. Semantic Fluency in nfvPPA (n = 518) and Phonemic Fluency in lvPPA (n = 208) were the most promising for these subtypes. A prior study of nfvPPA showed a similar estimate (n = 507) to detect the same treatment effect with semantic fluency. Sum of Boxes was among the best measures for all groups, consistent with prior work. ,

There are several potential explanations for the difficulty differentiating nfvPPA and lvPPA, and the high sample size estimates needed to detect meaningful change in these variants. For example, the lack of differentiation of the nfvPPA group from others on the NAT could result from the fact that the nfvPPA category can include individuals with either grammatical processing deficits or motor speech deficits. , Although the NAT was designed to avoid speech output problems, some non‐fluent/agrammatic cases may lack agrammatism. Furthermore, our finding that only the lvPPA group performed lower than the bvFTD group on the NAT is difficult to interpret. It may indicate that the NAT is sensitive to their known working memory deficits.

Many FTLD‐MOD measures, although theoretically sound, have psychometric problems. Notably, controls performed at ceiling on most measures. In addition, several measures have only a few items, which reduces the reliability and variance of the tests. For example, repetition tasks are commonly used clinically to distinguish lvPPA from other diagnoses, but the FTLD‐MOD repetition task includes only five items of familiar content. Recent work using a 20‐item repetition task, with phrases varying in length, meaningfulness, and familiarity, evidenced 89% accuracy for classifying PPA subtypes, compared to 67% in the current study. Future studies should seek to evaluate repetition tasks with more items, longer, less frequent sentence structures, and greater phonemic complexity.

The ability of a single task to discriminate among clinical syndromes is a high bar and does not map directly to clinical diagnostic practices. In the clinic, neuropsychological tests are only one component of a clinician's diagnostic armamentarium and are not considered in isolation. In this research study; however, we wanted to limit circularity and focus on FTLD‐MOD tasks, which were not used for syndromic diagnoses. Future studies could explore the utility of adding FTLD‐MOD tasks to other aspects of the clinical workup. A second limitation is that we focused on total scores for these tasks. The nature of language disorders, however, is that poor performance on any given test could be the consequence of several different language impairments. Future work should include a granular analysis of additional test variables, such as types of errors, that might help improve diagnostic distinctions. Although one of this study's strengths was the large sample size, there are important aspects of phenotyping (eg, identifying familial cases) that were not conducted in this sample. Finally, the samples that completed UDS measures (collection started in March 2015) were often smaller than those that completed the FTLD‐MOD, and thus results warrant replication.

In summary, the FTLD‐MOD tasks perform well at distinguishing PPA from bvFTD and distinguishing svPPA from other PPA subtypes. UDS tasks perform best at discriminating among all PPA subtypes. Longitudinal results paralleled the cross‐sectional findings and suggested that naming tests are potentially useful as trial endpoints. SvPPA stands out as a strong indication for treatment trials, given the accuracy of diagnosis, the array of sensitive outcomes, and strong clinicopathological correlation. These results may help clinicians with their test selection and highlight the need for developing and validating better endpoints for disease monitoring in lvPPA and nfvPPA.

CONFLICT OF INTEREST

BB has served as an investigator for clinical trials sponsored by Axovant and Biogen. He receives royalties from the publication of a book entitled Behavioral Neurology of Dementia (Cambridge Medicine, 2009, 2017). He serves on the Scientific Advisory Board of the Tau Consortium. He receives research support from the NIH, the Mayo Clinic Dorothy and Harry T. Mangurian Jr. Lewy Body Dementia Program, and the Little Family Foundation. HJR has a consulting agreement Ionis Pharmaceuticals. JHK has provided consultation for Biogen. He helped develop the Delis–Kaplan Executive Function System and receives royalties from Pearson Education, Inc., for helping to develop the California Verbal Learning Test. ALB receives research support from NIH, the Tau Research Consortium, the Association for Frontotemporal Degeneration, Bluefield Project to Cure Frontotemporal Dementia, Corticobasal Degeneration Solutions, the Alzheimer's Drug Discovery Foundation, and the Alzheimer's Association. He has served as a consultant for Aeton, Abbvie, Alector, Amgen, Arkuda, Ionis, Iperian, Janssen, Merck, Novartis, Samumed, Toyama, and UCB; and has received research support from Avid, Biogen, BMS, C2N, Cortice, Eli Lilly, Forum, Genentech, Janssen, Novartis, Pfizer, Roche, and TauRx.

Supplementary information

Click here for additional data file.