A Retrospective Study of Long-Term Improvement on the Boston Naming Test

Abstract

Purpose

Lexical retrieval impairment is a universal characteristic of aphasia and a common treatment focus. Although naming improvement is well documented, there is limited information to shape expectations regarding long-term recovery. This was the motivation for a retrospective study of longitudinal data on the Boston Naming Test (BNT; Kaplan, Goodglass, & Weintraub, 1983, 2000).

Method

BNT scores were analyzed from a heterogeneous cohort of 42 individuals with anomia associated with a range of aphasia types. The data were collected over the course of 20 years from individuals who had participated in treatment and received at least 2 BNT administrations. A linear mixed model was implemented to evaluate effects of initial BNT score, time postonset, and demographic variables. For those over 55 years of age, BNT change was evaluated relative to data from the Mayo Clinic's Older Americans Normative Studies.

Results

There was a significant average improvement of +7.67 points on the BNT in individuals followed for an average of 2 years. Overall, the average rate of improvement was +5.84 points per year, in contrast to a decline of 0.23 points per year in a healthy adult cohort from the Mayo Clinic's Older Americans Normative Studies. Naming recovery was approximately linear, with significant main effects of initial BNT score (i.e., initial severity) and time postonset; the greatest changes were noted in those whose initial severity was moderate.

Conclusions

These findings indicate a positive prognosis for naming improvement over time regardless of demographic factors and provide estimates for clinical predictions for those who seek rehabilitation during the chronic phase.

Word retrieval difficulties are a universal characteristic of aphasia, regardless of aphasia type or severity (Benson, 1979; Goodglass, 1993). In fact, naming problems are often the primary complaint of individuals with aphasia and may persist for years following stroke. It is not surprising that lexical retrieval is the focus of a large corpus of aphasia treatment studies. Overall, there is substantial evidence of positive response to behavioral intervention, so there is an expectation of improved naming over time (for reviews, see Nickels, 2002; Raymer & Gonzalez Rothi, 2008). The majority of treatment studies were designed to examine the therapeutic value of a given approach using single-subject designs conducted over relatively short periods of time (weeks or months). Such data have promoted the development and refinement of intervention strategies, but they offer relatively little information regarding the long-term recovery of naming impairment.

In the broader context, lexical retrieval skills are often a component of group studies designed to examine overall language recovery (e.g., Aftonomos, Steele, & Wertz, 1997; El Hachioui et al., 2013; Kertesz & McCabe, 1977; Laska, Hellblom, Murray, Kahan, & Von Arbin, 2001; Nicholas, Helm-Estabrooks, Ward-Lonergan, & Morgan, 1993; Shewan & Kertesz, 1984; Wertz et al., 1981). Collectively, such studies demonstrate language improvement well into the chronic stage, but there is limited information to shape expectations regarding the magnitude of improvement or the rate of change that might be expected in the years following onset of aphasia. Relevant to this issue, Plowman, Hentz, and Ellis (2012) conducted a comprehensive review of aphasia recovery literature to examine the influence of patient- and stroke-related factors on long-term outcomes. They found that stroke-related factors captured by measures of initial severity were among the strongest predictors of language outcome, whereas the predictive value of patient-related factors, such as age, education, and gender, was limited. At an individual level, stroke-related factors would include the location and extent of brain damage, as well as the consequent status of the cognitive processes and sensorimotor abilities that support language. In the context of most group studies, standardized measures of language performance are the most consistent estimate of initial severity and long-term outcomes (e.g., the Aphasia Quotient from the Western Aphasia Battery [WAB]; Kertesz, 1982). In the case of naming ability, performance on the Boston Naming Test (BNT; Kaplan, Goodglass, & Weintraub, 1983, 2000) is the most common measure.

A review of the language recovery literature yielded very few studies that provide insight regarding long-term recovery of naming ability in those with aphasia. Aftonomos et al. (1997) examined language recovery in 23 individuals with chronic aphasia following completion of a treatment program using the Lingraphica device. These participants were followed for about 16 weeks, and overall, the researchers documented significant improvement across a range of aphasia types and severities. For 10 of the 23 participants, data were available on the BNT over time, showing an average improvement of +11.1 points in this cohort who ranged in time postonset (TPO) of aphasia from 6 months to 6 years and had an average initial BNT score of 23.0 (range = 0–40). This finding demonstrates potential for significant improvement in naming ability during the chronic stage, particularly with specific intervention.

Recognizing the limited data regarding long-term outcomes in aphasia recovery, Holland, Fromm, Forbes, and MacWhinney (2017) took advantage of the AphasiaBank database (MacWhinney, Fromm, Forbes, & Holland, 2011) to examine repeated standardized test data collected from 26 individuals over at least 1 year. They documented significant improvement on the Western Aphasia Battery–Revised (Kertesz, 2006) in a subgroup of 16 (mean TPO = 4.9 years) who were followed for an average of 3.9 years. Eleven of those people were also tested on the short form of the BNT, and they improved from an average of 7.5–9.1 on the 15-item test. These individuals represented diverse demographic backgrounds and participated in a variety of community aphasia programs at different locations.

Similar to Holland et al. (2017), we sought to learn from a relatively large set of data that had been collected at the University of Arizona over the past 20 years. The University of Arizona has a history of providing services to individuals with aphasia and developing and evaluating treatment methods for various associated impairments, including anomia. The BNT was frequently administered during initial evaluations and periodic reassessments over the course of time that often extended several years. We anticipated naming abilities would improve over time and sought to determine what magnitude of change might be expected, as well as identify factors with predictive value in estimating long-term recovery. Given that our cohort included many older adults, however, we questioned whether changes over time might be influenced by age-related effects apart from the stroke. Indeed, this was a relevant concern as we consider naming performance, because there is evidence of a significant decline in naming ability after 55 years of age in healthy older adults (Feyereisen, 1997; Ivnik, Malec, Smith, Tangalos, & Petersen, 1996; Mitrushina, Boone, Razani, & D'Elia, 2005).

The age-related decline in naming was clearly documented in the context of the extensive Mayo Clinic's Older Americans Normative Studies (MOANS), which included administration of the 60-item BNT to 663 cognitively healthy older adults aged 56–99 years (Ivnik et al., 1996). Ivnik et al. found that age explained approximately 21% (r² = .21) of the raw score variance in BNT scores. By contrast, sex and education had minimal influence on BNT performance in this older cohort, with sex explaining approximately 4% (r ² = .04) of the raw score variance and education explaining approximately 7% (r ² = .07). Although the MOANS data were cross-sectional, rather than longitudinal, the decline of naming with age was evident. For example, when comparing the range of scores at the 50th percentile for adults in the 56- to 62-year age band to the 84- to 86-year range, raw scores on the BNT dropped from 55–56 to 49–52 correct. Considered relative to older adults with aphasia, this suggests that, as they work to recover from acquired language impairment, they may also experience age-related effects distinct from the cause of the aphasia. In order to address this issue in our study, we incorporated relevant BNT data from the MOANS project that was generously provided by Robert Ivnik.

Method

Sources of Data

A review of clinical and research records at the University of Arizona served to identify 62 adults (> 21 years) with acquired language impairment due to acute neurological damage who had participated in treatment and received at least two complete administrations of the 60-item BNT. This primarily included individuals who had experienced left middle or posterior cerebral artery strokes and not those with primary progressive aphasia, multi-infarct dementia, or complicated or unclear etiologies. From this data set, we excluded four people who did not remain neurologically stable over the time when data were collected due to subsequent neurological damage (e.g., additional strokes) or progressive cognitive decline suggesting the onset of dementia. We eliminated one person because his BNT score dropped 5 points over the course of a year, suggesting a significant decline according to criteria reported by Sachs et al. (2012). For the remaining 57 individuals, the times postonset of the initial BNT administrations were reviewed and restricted to those collected after 3 months. This identified data points from four people, but three of them had at least two BNT scores after the 3-month threshold, so their data were retained. To ensure the presence of naming impairment with room for improvement, we further restricted the data set to those with initial BNT scores between 0 and 51, causing the exclusion of 10 people with an initial BNT score of 52 or greater. Finally, we retained only those individuals who were followed for 0.40 years (4.8 months) or longer, which eliminated four who were seen for shorter periods of time.

After inclusion and exclusion criteria were applied, data from 42 individuals with aphasia were available for analysis. The average TPO associated with the initial BNT scores examined in this study was 1.83 years (SD = 2.02), ranging from 3 months to 8.11 years postonset. The number of BNTs administered to each individual ranged from two to eight tests (M = 3), which provided a data set of 126 BNT scores.

Demographic characteristics are summarized in Table 1, and individual data are included in the Appendix. The cohort included more men than women (29 vs. 13), and the average age was 60.28 years. All participants had at least 10 years of formal education (or a high school equivalency diploma), and half of the cohort had some post–secondary education; the average education for the entire cohort was 14.50 years. The range of aphasia severity as indexed by the WAB Aphasia Quotient was 16.7–96.1, and the aphasia profiles included anomic (21), Broca's (10), conduction (seven), Wernicke's (three), and global (one). Speech samples from the WAB were evaluated for evidence of apraxia of speech (AOS) using a 5-point apraxia rating scale, where 0 = no evidence of AOS, 1 = mild AOS, 2 = moderate AOS, 3 = marked AOS, and 4 = severe AOS. These ratings were operationalized using the Apraxia of Speech Rating Scale scoring instructions from Strand, Duffy, Clark, and Josephs (2014). Fourteen of the 42 participants had some degree of AOS, ranging from mild to severe (see Appendix).

Table 1.

Summary characteristics of 42 individuals with aphasia.

Characteristics	n	M	SD	Range
Gender: male/female	29/13
Age at initial BNT (years)	42	60.28	15.22	23.22–85.25
Education (years)	42	14.50	2.92	10–20
Initial Aphasia Quotient (of 100)	42	67.30	24.20	16.7–96.1
Initial BNT raw score (of 60)	42	24.07	18.38	0–51
TPO at initial BNT (years)	42	1.83	2.02	0.25–8.11
Time between first and last BNT (years)	42	2.09	2.01	0.41–7.38
Estimated treatment hours (individual)	42	39.24	18.30	15–74

Note. BNT = Boston Naming Test; TPO = time postonset; SD = standard deviation.

Test administration of the BNT had been accomplished in one of two contexts: either as part of a research protocol or within clinical service delivery. In all instances, it was performed by a certified speech-language pathologist or by a graduate student in the presence of a certified speech-language pathologist. Because the data were collected over more than a decade, there were a variety of testers who all followed the standard test instructions. Data integrity had been checked at several time points, including when initial evaluation reports were generated, when case presentations were given, and as data were aggregated for this retrospective analysis. The majority of test administrations were video-recorded so that any ambiguous scores could be reevaluated. Sums were checked on the hard copy of the response forms to assure correct totals.

The initial BNT scores were distributed throughout the range from 0 to 51 out of 60 points, and the age at the time of the first BNT administration ranged widely from 23.22 to 85.25 years (see Figure 1). Each score was evaluated relative to age-adjusted norms using Ivnik et al. (1996) for those over 55 years and Tombaugh and Hubiey (1997) for 55 years and younger. Most scores fell below the 10th percentile, with the highest score at the 29th percentile.

Figure 1.

Initial Boston Naming Test (BNT) scores relative to age at initial test for individuals with aphasia (black circles) and average scores from the Mayo Clinic's Older Americans Normative Studies (open circles with standard error bars).

A review of records confirmed all individuals with aphasia were enrolled in therapy during one or more time periods when the BNT data were collected, but they were not always continuously enrolled in individual and/or group treatment. All participants had received treatment that focused on the nature of their language impairment, which included lexical retrieval treatment in some, but not all, cases. Treatment was administered in an individual context for all but one person who had received lexical retrieval treatment in a group context only. We quantified treatment by the number of direct hours of therapy, which ranged from 15 to 74, with an average of 39.17 total hours (SD = 18.30). In most cases when the treatment total was greater than 20 hr, individuals had received a sequence of treatments in a research context to incrementally advance language performance. We could also document that 21 of the 42 individuals participated in aphasia group sessions in our clinic for some time (see Appendix), but the number of hours of group participation was not quantified because reliable information was not available for all participants.

To consider change in naming performance in our cohort with aphasia relative to the normative BNT data from the MOANS (Ivnik et al., 1996), we performed analyses using raw BNT data for 663 healthy older adults who participated in the MOANS. We specifically looked at individuals in the 56- to 85-year age range (eliminating those older than 85 years) and included people who more closely matched the age and education of our participants. This yielded data from 400 people who we could compare to the older individuals in our cohort, a subgroup of 28 individuals with aphasia between 56 and 85 years at the time of their first BNT. Demographic information for these subcohorts is included in Table 2. The average age of the MOANS participants was 70.24 years (SD = 8.43), with an average education of 14.45 years, which did not differ significantly from our older subgroup (28 individuals) for age, t(31.87) = −0.66, p = .513, or education, t(28.88) = 0.48, p = .632. The average BNT score for the MOANS cohort was 53.69 (SD = 5.17), and the average initial BNT for the older subgroup with aphasia was 23.71 (SD = 17.16). For visual comparison, Figure 1 includes the average BNT scores for 400 healthy adults from the MOANS data set shown in 5-year increments.

Table 2.

Demographic information and Boston Naming Test (BNT) performance for older individuals with aphasia and healthy adults from the Mayo Clinic's Older Americans Normative Studies (MOANS).

Aphasia	n = 28	M	SD	Range
Gender: male/female	20/8
Age at initial BNT (years)		69.25	7.58	58.0–85.2
Education (years)		14.75	3.34	10–20
TPO at initial BNT (years)		1.85	2.03	0.25–8.11
Initial BNT (of 60)		23.71	17.16	1–49
Time between first and last BNTs (years)		1.88	1.87	0.41–7.34
MOANS a	n = 400	M	SD	Range
Age at test (years)		70.24	8.43	56–85
Education (years)		14.45	2.26	10–20
BNT (of 60)		53.69	5.17	25–60

Note. TPO = time postonset; SD = standard deviation.

^a Male/female not distinguished.

Statistical Analysis

To characterize overall change in BNT performance, the initial and final BNT scores were compared using a random effects linear mixed model implemented in SPSS Version 24, with test time and TPO as fixed factors and participant as the random factor. To examine all BNT scores over time, a linear mixed model was implemented with initial BNT scores and the TPO as fixed factors, and all subsequent BNT scores (excluding initial BNT) as the dependent measure. In order to model individual variability with regard to initial severity and change over time, random intercepts and random slopes were also included in the model (Model 1). We compared this initial model to several others: a model containing a quadratic term for TPO to examine the linearity of slope of recovery (Model 2), a model with fixed effects for demographic variables (age at stroke, education, gender) to determine whether demographic characteristics were significant predictors of BNT performance over time (Model 3), and a model with a fixed effect for the number of individual treatment hours to evaluate the influence of treatment on BNT scores (Model 4).

We also calculated the slope of the BNT performance relative to age from the MOANS data. Recognizing that the MOANS data were cross-sectional rather than longitudinal, we used the BNT performance relative to age as a proxy for age-related change that might be expected in older adults over time.

Results

Figure 2 shows the BNT scores for each person with aphasia plotted against the respective times postonset in years. This visual display of individual performance shows the change in BNT scores (ranging from −1 to +25 points) over the time followed (ranging from about 5 months to 7.5 years). Overall, the average initial BNT score was 24.07 (SD = 18.38), and the average final BNT score was 31.74 (SD = 18.64), yielding a mean change of +7.67 over an average of 2.09 years. The final BNT scores for five individuals fell above the 30th percentile relative to age-adjusted norms, with raw scores between 54 and 57 out of 60. The linear mixed-model analysis showed the overall change in BNT was significant, β = 5.53, t = 3.95, p < .001, with a significant influence of TPO, β = 1.02, t = 2.19, p = .033.

Figure 2.

Boston Naming Test (BNT) scores in relation to time postonset of aphasia for 42 individuals.

The results of the mixed-model analyses designed to examine change in BNT scores sampled over various times postonset are summarized in Table 3. There was a significant main effect of initial severity as indicated by initial BNT score, β = 0.95, t = 17.62, p < .001, and TPO in years on subsequent BNT scores, β = 1.08, t = 2.88, p = .008. Individual variability in intercepts accounted for 70% of the total variance in BNT scores, while random slopes accounted for less than 3%. Recall that scores were predicted from the initial BNT, so the intercept and slope for fixed effects were modeled for all subsequent BNT scores. This means that changes from the first to the second test administration were captured by both the intercept and the coefficient associated with initial BNT scores, and subsequent change over time was modeled by the slope across time (TPO). In Model 2, the quadratic term for TPO was not significant, indicating a linear relation between BNT scores and TPO. In Model 3, the demographic variables (age at stroke, education, gender) were not statistically significant, and in Model 4, number of individual treatment hours did not have a significant effect.

Table 3.

Mixed-model results with Boston Naming Test (BNT) raw score as the dependent variable.

Model 1: Examining initial BNT score and TPO
	b	SE	t	p	95% CI
Intercept	4.38	1.93	2.27	.028	[0.50, 8.27]
Initial BNT	0.95	0.05	17.62	< .001	[0.84, 1.06]
TPO in years	1.08	0.38	2.88	.008	[0.31, 1.86]
Component	% Variance	SE	p
Intercept	0.70	9.54	.009
Slope	0.02	0.51	.095
Residual	0.27	2.53	< .001
−2 Log likelihood = 516.40 No. of parameters = 6
Model 2: Examining linearity of recovery slope of BNT performance
	b	SE	t	p	95% CI
Intercept	4.99	2.25	2.21	.031	[0.47, 9.50]
Initial BNT	0.95	0.05	17.66	< .001	[0.84, 1.06]
TPO in years	0.77	0.75	1.02	.311	[−0.73, 2.26]
TPO2	0.03	0.06	0.48	.634	[−0.10, 0.16]
Component	% Variance	SE	p
Intercept	0.70	9.49	.012
Slope	0.03	0.53	.093
Residual	0.28	2.52	< .001
−2 Log likelihood = 516.19 No. of parameters = 7
Model 3: Examining the influence of demographic variables on BNT
	b	SE	t	p	95% CI
Intercept	4.09	6.04	0.68	.503	[−8.18, 16.36]
Initial BNT	0.94	0.06	17.01	< .001	[0.83, 1.06]
TPO in years	1.11	0.38	2.95	.007	[0.33, 1.88]
Age at stroke	0.03	0.07	0.49	.625	[−0.10, 0.17]
Education	−0.12	0.36	−0.34	.739	[−0.85, 0.61]
Gender	0.26	2.29	0.11	.911	[−4.39, 4.91]
Component	% Variance	SE	p
Intercept	0.71	9.64	.009
Slope	0.02	0.50	.118
Residual	0.27	2.53	< .001
−2 Log likelihood = 516.11 No. of parameters = 9
Model 4: Examining the influence of treatment
	b	SE	t	p	95% CI
Intercept	3.65	3.36	1.09	.285	[−3.18, 10.49]
Initial BNT	0.96	0.06	16.16	< .001	[0.84, 1.08]
TPO in years	1.07	0.38	2.81	.010	[0.28, 1.85]
Hours of treatment	−0.02	0.06	0.27	.789	[−0.10, 0.14]
Component	% Variance	SE	p
Intercept	0.70	9.52	.010
Slope	0.02	0.51	.093
Residual	0.27	2.52	< .001
−2 Log likelihood = 516.33 No. of parameters = 7

Note. TPO = time postonset; SE = standard error; CI = confidence interval.

Although the random effect for slope was minimal, visual inspection of the recovery curves in Figure 2 suggested greater change for individuals in the mid-range with regard to initial severity. In contrast, it appeared there was more modest improvement for individuals with marked naming impairment at the outset and those with impairments in the mild range. To further explore these apparent differences in trajectory of recovery based on initial severity, we divided the cohort into three groups based on initial BNT scores. Taking advantage of breaks in the distribution of initial scores in our cohort, we designated those with scores of 0–8 as severe (n = 13), 10–35 as moderate (n = 16), and 41–51 as mild (n = 13; see Table 4). The group differences in severity were significant, F(2, 39) = 163.61, p < .001, and Tukey's post hoc comparisons confirmed each group was different from the others, p < .05. The overall change by initial severity was +5.15 for severe, +12.31 for moderate, and +4.46 for mild, with the moderate group showing significantly greater change than the severe and mild groups, F(2, 39) = 7.63, p = .002, Tukey's p < .05. The groups did not differ with regard to age, education, TPO, or the length of time they were followed. However, there was a significant difference regarding the number of hours of treatment, F(2, 39) = 4.04, p = .026, with the severe group receiving more treatment than the mild group, Tukey's p < .05.

Table 4.

Characteristics of subgroups based on initial Boston Naming Test (BNT) score.

Severe	n = 13	M	SD	Range
Gender: male/female	11/2
Age at stroke (years)		51.29	16.13	22.48–70.29
Education (years)		15.08	2.53	12–20
Time postonset at initial test		2.94	2.90	0.38–8.11
Initial BNT score (of 60) a		3.54	2.88	0–8
Absolute BNT change b		5.15	5.63	0–17
Time between first and last BNTs (years)		1.70	1.43	0.48–6.12
No. of treatment hours c		48.08	18.50	24–74
Moderate	n = 16	M	SD	Range
Gender: male/female	7/9
Age at stroke (years)		64.01	14.84	35.28–84.93
Education (years)		13.81	3.35	10–20
Time postonset at initial test		1.20	1.08	0.25–4.55
Initial BNT score (of 60) a		22.13	9.24	10–35
Absolute BNT change b		12.31	8.03	1–25
Time between first and last BNTs (years)		2.32	2.29	0.65–7.34
No. of treatment hours c		40.25	18.26	15–71
Mild	n = 13	M	SD	Range
Gender: male/female	11/2
Age at stroke (years)		58.67	14.05	36.73–76.70
Education (years)		14.77	2.77	12–20
Time postonset at initial test		1.49	1.44	0.28–4.64
Initial BNT score (of 60) a		47.00	2.83	41–51
Absolute BNT change b		4.46	2.70	−1–9
Time between first and last BNTs (years)		2.20	2.21	0.41–7.38
No. of treatment hours c		29.15	13.74	15–53

Note. SD = standard deviation.

^a Groups significantly differ on initial BNT score.

^b Groups significantly differ on BNT change (moderate > mild and severe).

^c Groups significantly differ on number of treatment hours (severe > mild).

We examined differences between the three severity groups using a mixed model containing a fixed effect for severity group (mild, moderate, severe) and an interaction term for severity and TPO (see Table 5). The main effect of severity group was not statistically significant, but the interaction term for severity and TPO was significant, β = 1.90, t = 3.04, p < .003. This confirmed a significant difference among severity groups with respect to time, indicating that separate prediction equations for BNT scores over time would better capture the trajectory of recovery based on initial severity. Thus, we proceeded to run a separate linear mixed model for each severity group. Each model (mild, moderate, severe) contained a fixed effect for initial BNT score and TPO at the time of each test, and the dependent measure included all subsequent BNT scores (excluding initial). As shown in Table 5, there was a significant main effect for initial BNT score for the severe and moderate groups, and TPO was a significant predictor for the moderate group. Using the results from the mixed models, we derived prediction equations for each severity group, and a visual representation of the recovery curves is depicted in Figure 3. As illustrated in this figure, the data indicate that individuals of all levels of severity are likely to improve over time, but those of moderate severity show greater improvement with the passage of time.

Table 5.

Mixed-model results by severity group with prediction equations.

Group	b	SE	t	p	95% CI
Severe
Intercept	2.68	2.28	1.18	.266	[−2.38, 7.75]
Initial BNT	1.54	0.62	2.48	.024	[0.23, 2.85]
TPO in years	0.14	0.43	0.31	.757	[−0.77, 1.04]
Predicted BNT = 2.68 + (1.53 × Initial BNT) + (0.14 × TPO)
Moderate
Intercept	6.39	5.26	1.21	.243	[−4.78, 17.55]
Initial BNT	0.84	0.20	4.11	.001	[0.40, 1.28]
TPO in years	2.39	0.48	4.98	< .001	[1.42, 3.37]
Predicted BNT = 6.39 + (0.84 × Initial BNT) + (2.39 × TPO)
Mild
Intercept	20.69	13.41	1.54	.145	[−8.08, 49.46]
Initial BNT	0.61	0.29	2.14	.050	[−0.0009, 1.23]
TPO in years	0.39	0.26	1.54	.138	[−0.14, 0.93]
Predicted BNT = 20.69 + (0.61 × Initial BNT) + (0.39 × TPO)

Note. BNT = Boston Naming Test; TPO = time postonset; SE = standard error; CI = confidence interval.

Figure 3.

Predicted Boston Naming Test scores based on equations derived from mixed-model analyses. Prediction equations by severity group (BNT = Boston Naming Test; TPO = time postonset): Mild = 20.69 + (0.61 × Initial BNT) + (0.39 × TPO). Moderate = 6.39 + (0.84 × Initial BNT) + (2.39 × TPO). Severe = 2.68 + (1.54 × Initial BNT) + (0.14 × TPO).

The rate of change on the BNT over time was calculated for each individual by dividing the amount of change in BNT points by the time followed (in years). This yielded an average improvement rate of +5.84 (SD = 6.92) points per year for those with aphasia. By contrast, the data from the 400 healthy adults in the MOANS revealed a decline of 0.23 points on the BNT per year. When we restricted scores to our older cohort with aphasia (56–85 years, n = 28), they showed an average gain of 6.79 points per year (SD = 7.83). These findings demonstrate the recovery of naming skills was robust to potential age-related decline in the older cohort.

Discussion

We report here on a retrospective study of long-term performance on the BNT in a cohort of 42 individuals who participated in treatment in a university setting, with data collected over the course of 20 years. The mean overall gain on the BNT was +7.67 points over an average interval of about 2 years, and the average rate of change was +5.84 points per year across the entire group. Initial performance on the BNT and the passage of time were the best predictors of improvement over time, whereas demographic characteristics, such as education, gender, or age at time of stroke, did not have a significant influence on BNT performance. The limited predictive value of demographic variables was consistent with other studies (e.g., Pedersen, Vinter, & Olsen, 2004; Plowman et al., 2012). We also documented that the average rate of improvement of the older participants in our study (> 55 years) was over 6 points per year, which far outweighed the small but meaningful decline of 0.23 points documented in the group of healthy adults from the MOANS cohort. If it is appropriate to assume the older individuals with aphasia might also have been fighting some age-related decline, then the estimated “net” improvement would be slightly higher. These findings are helpful to demonstrate the effects of aging are quite small relative to long-term recovery from acquired language impairment.

When considered relative to initial severity, the greatest gains were made by those in the moderate group. This subgroup of 16 individuals with initial scores ranging from 10 to 35 (out of 60) were comparable to those studied by Aftonomos et al. (1997), who improved an average of 11 points in comparison to 12 points from our group. To compare our data to the retrospective study by Holland et al. (2017), who used AphasiaBank data, we converted raw scores to percent correct. In their subgroup who showed improvement, the average was 1.6 of 15 items (10.6%), which was comparable to an average improvement of 13% (7.67 out of 60) in our full cohort. If we consider the improvement of our moderate group alone, there was a 21% gain (12.31 out of 60). These comparisons show that our results appear consistent or stronger in terms of magnitude of change observed by others. This included relatively dramatic improvement in some individuals, namely, the four in the moderate group who gained 20 or more points on the 60-item test.

The cohort with the most severe naming impairment demonstrated an average improvement of about 5 points, but they did not show a continuation of improvement over time. This group of 13 included five individuals with marked AOS as well as three individuals with Wernicke's aphasia and several with conduction aphasia who had persistent problems with phonological assembly resulting in paraphasic errors. As expected, impairment to these more peripheral processes tends to limit the ability to improve spoken naming. Seven of the 13 individuals with severe impairment at the outset showed limited improvement (+2 or fewer points) on the BNT, but it was interesting to note that four individuals in this group improved by 10 or more points, and two of them had moderately severe AOS (see Appendix). Thus, although factors that contributed to poor naming at the outset were likely to limit the extent of improvement in many cases, our findings show some individuals can certainly exceed expectations.

It was not surprising that the group with the milder naming impairment demonstrated the most modest gains: an average of 4.5 points on the BNT. The overall improvement was likely influenced by a “ceiling effect.” In fact, five individuals reached scores within the normal range of their age-matched peers. As might be expected, the individuals in this group all had profiles consistent with anomic aphasia from the outset. As noted in the other two groups, some individuals made sizable improvements of 8 or 9 points.

Given that the initial performance on the BNT and the passage of time were the strongest predictors of naming ability over time, these values might be used to predict the outcomes for a given individual. To facilitate this, we generated a table using the slope equations derived for three broad levels of initial severity on the BNT. Table 6 provides predicted outcomes for representative BNT scores and expected change over time. By way of example, a clinician with a patient who was 1 year postonset with a BNT score of 15 would predict from the table that, in 1 year, the performance might be about 21, and in the following year, it would be about 24, and so on. Such information can help set expectations about the potential for continued improvement or to discern whether progress appears to be on par (or above or below) expectations. The values must be used with caution, however, because there is residual error associated with the predictions, and the treatment characteristics were not controlled in this retrospective study. In our cohort, treatment averaged about 20 hr a year, but the amount of treatment did not predict the magnitude of gain on the BNT. This is not so surprising because the amount of treatment may be more of a reflection of severity. In our study, those with more severe naming impairment received more treatment than those with mild impairment. Inferences about treatment were also limited because it was not possible to parse potential contributions from group therapy. With these caveats in mind, the findings still have considerable value as estimates of potential improvement.

Table 6.

Predicted Boston Naming Test (BNT) scores based on initial severity and time postonset (TPO).

Severity Level	Out of 60	Years later	Predictions
BNT range	Initial BNT	TPO	Predicted BNT	Cumulative gain
Severe 0–8	1	1	4.36	3.36
	1	2	4.49	3.49
	1	3	4.63	3.63
	1	4	4.77	3.77
	1	5	4.90	3.90
	5	1	10.50	5.50
	5	2	10.64	5.64
	5	3	10.78	5.78
	5	4	10.91	5.91
	5	5	11.05	6.05
	8	1	15.11	7.11
	8	2	15.25	7.25
	8	3	15.38	7.38
	8	4	15.52	7.52
	8	5	15.66	7.66
Moderate 10–35	10	1	17.19	7.19
	10	2	19.58	9.58
	10	3	21.97	11.97
	10	4	24.37	14.37
	10	5	26.76	16.76
	15	1	21.39	6.39
	15	2	23.78	8.78
	15	3	26.18	11.18
	15	4	28.57	13.57
	15	5	30.96	15.96
	20	1	25.59	5.59
	20	2	27.98	7.98
	20	3	30.38	10.38
	20	4	32.77	12.77
	20	5	35.16	15.16
	25	1	29.79	4.79
	25	2	32.19	7.19
	25	3	34.58	9.58
	25	4	36.97	11.97
	25	5	39.37	14.37
	30	1	33.99	3.99
	30	2	36.39	6.39
	30	3	38.78	8.78
	30	4	41.17	11.17
	30	5	43.57	13.57
	35	1	38.20	3.20
	35	2	40.59	5.59
	35	3	42.98	7.98
	35	4	45.38	10.38
	35	5	47.77	12.77
Mild 40 and up	40	1	45.63	5.63
	40	2	46.03	6.03
	40	3	46.42	6.42
	40	4	46.82	6.82
	40	5	47.21	7.21
	45	1	48.70	3.70
	45	2	49.10	4.10
	45	3	49.49	4.49
	45	4	49.89	4.89
	45	5	50.28	5.28
	50	1	51.77	1.77
	50	2	52.17	2.17
	50	3	52.56	2.56
	50	4	52.96	2.96
	50	5	53.35	3.35
	55	1	54.84	−0.16
	55	2	55.24	0.24
	55	3	55.63	0.63
	55	4	56.03	1.03
	55	5	56.42	1.42

In our view, the documented improvements on the BNT were impressive for a number of reasons. First, the BNT is a relatively challenging confrontation naming test that examines performance on untrained items that includes those in the low-frequency range. Even when naming is the specific focus of behavioral intervention, improvement on the BNT indicates benefit beyond a direct treatment effect. Although these data demonstrate a significant, generalized improvement in naming ability, the retrospective nature of this study did not allow a consistent measure of the functional value of this improvement in our cohort. The documented relation between improved picture naming and improved lexical retrieval in connected speech provided by others (e.g., Conroy, Sage, & Lambon Ralph, 2009; Herbert, Hickin, Howard, Osborne, & Best, 2008) supports our expectation that the improved performance on the BNT should translate to greater success in word retrieval in everyday conversation. This expectation is consistent with the widespread use of confrontation naming tests as a proxy for word retrieval abilities in real life (Laine & Martin, 2006; Lezak, Howieson, & Loring, 2004).

The current study appears to be the largest to examine long-term recovery of naming abilities in individuals with chronic aphasia, but we acknowledge the size is small relative to an ideal cohort representing a broader range of times postonset and aphasia severity. We also note that BNT assessments were not always time-locked to the end of a treatment protocol, which may have reduced the robustness of treatment outcomes in some instances, but it also offers credence to the apparent durability of improvements over time. The use of initial severity and time fails to capture important information regarding the status of underlying cognitive skills, such as semantic knowledge and phonological skills, that are known to predict naming performance and response to treatment (e.g., see El Hachioui et al., 2013; Lambon Ralph, Moriarty, & Sage, 2002; Lambon Ralph, Snell, Fillingham, Conroy, & Sage, 2010). These factors are important for the design and implementation of optimal treatment and were certainly considered in the treatments implemented in our cohort. We also did not capture variations in person-specific factors such as motivation or family support; however, all participants had the common history of seeking rehabilitation services in the years following the onset of aphasia. They also had the motivation to attend treatment sessions and were either independent enough or had adequate family support to manage the consistent travel to the university. These within-person factors are not trivial and are among those likely to be discernable in typical clinical settings.

In summary, our data reflect the findings from a heterogeneous cohort of individuals with aphasia during the chronic phase of recovery. The limited effect of demographic variables on outcomes is good news for individuals with aphasia in that the expectation of continued recovery is not tempered by variables that cannot be changed. A university clinic and research lab offer an ideal context to promote continued language recovery, but the translation to other clinical settings and community aphasia centers is important as well. Given that cost and reimbursement issues clearly influence opportunities for long-term rehabilitation, the data presented here should be of value to demonstrate the potential benefit of treatment in chronic aphasia. We acknowledge these data only provide insight regarding performance on a challenging, structured naming task. While we expect improved performance on this test relates to better lexical retrieval in everyday context, we recognize many changes can occur with regard to functional communication that also significantly impact the quality of life.

Appendix

Demographic Characteristics of 42 Individuals With Aphasia

ID	CVA age	Age first test	Sex	Ed	WAB AQ	Aphasia type	Apraxia rating	Initial TPO (years)	Final TPO (years)	Time followed (years)	Initial BNT	Final BNT	Total BNT change	BNT change/year	No. of BNTs	Direct Tx hours
154	45.6	46.1	F	16	32.5	Wernicke's	0	0.52	1.34	0.82	0	1	1	1.21	2	30
162	29.4	30.9	M	12	27.6	Broca's	4	1.44	2.02	0.57	0	1	1	1.75	2	25 b
324	22.8	23.2	M	16	18.1	Global	3	0.38	1.42	1.04	0	10	10	9.63	3	55
322	57.3	59.6	M	14	16.7	Broca's	4	2.30	3.72	1.42	1	2	1	0.70	2	40 b
194	61.8	63.5	M	20	41.3	Wernicke's	0	1.62	2.68	1.06	2	2	0	0.00	3	24
131	62.8	63.6	M	18	53.9	Conduction	0	0.82	2.80	1.97	3	7	4	2.03	4	67 b
150	70.3	71.7	M	16	49.5	Conduction	0	1.35	2.66	1.31	3	5	2	1.52	2	73
187	57.2	61.5	M	16	29.3	Broca's	3	4.31	4.79	0.48	4	18	14	29.17	2	42
182	64.6	65.2	M	16	41	Broca's	3	0.61	2.08	1.48	5	7	2	1.35	2	57 b
159	22.5	29.5	M	12	51.7	Broca's	1	6.97	9.25	2.28	6	9	3	1.31	3	67 b
101	55.8	64.0	M	12	50.6	Conduction	0	8.11	9.82	1.71	7	9	2	1.17	2	28 b
176	58.4	66.4	F	16	54.7	Wernicke's	0	7.96	14.08	6.12	7	17	10	1.64	5	43
189	58.3	60.2	M	12	65.4	Conduction	2	1.80	3.65	1.85	8	25	17	9.19	2	74 b
196	57.5	58.0	F	12	37.9	Broca's	3	0.49	1.55	1.06	10	35	25	23.58	2	71 b
316	35.3	39.9	F	12	43.2	Broca's	3	4.55	5.38	0.83	10	12	2	2.40	2	62 b
007	41.7	42.2	M	12	90.9	Anomic	0	0.40	2.61	2.21	12	29	17	7.69	6	20 b
310	68.7	69.9	M	11	66.5	Conduction	0	1.08	3.16	2.07	12	23	11	5.30	3	68
161	51.0	53.2	F	12	36.6	Broca's	0	2.16	8.95	6.79	15	35	20	2.95	5	60 b
108	67.8	69.3	F	12	52.4	Conduction	1	1.54	8.88	7.34	16	40	24	3.27	5	33 b
130	74.0	74.6	M	18	89.6	Anomic	0	0.55	6.23	5.68	16	36	20	3.52	8	20 b
135	79.8	80.1	F	12	90.4	Anomic	0	0.28	0.94	0.66	20	23	3	4.53	3	15
308	50.7	51.1	M	18	67.1	Conduction	1	0.38	1.52	1.15	26	43	17	14.81	2	31 b
327	69.9	70.9	F	12	75.3	Anomic	0	0.89	2.25	1.37	26	39	13	9.51	4	48
009	84.9	85.2	M	20	92.5	Anomic	0	0.25	0.98	0.73	28	34	6	8.18	3	20
335	67.1	68.4	M	20	84.8	Anomic	0	1.25	1.90	0.65	28	37	9	13.74	2	36
158	75.6	77.8	F	12	83.5	Anomic	2	2.14	3.14	1.00	31	32	1	1.00	3	47 b
199	74.5	75.7	M	16	78	Broca's	3	1.12	2.19	1.07	34	35	1	0.93	2	51
137	47.4	48.7	F	12	77.5	Broca's	0	1.34	5.10	3.75	35	49	14	3.73	3	29 b
315	78.3	79.2	F	10	69.5	Anomic	0	0.81	1.54	0.73	35	49	14	19.07	3	33 b
123	76.7	78.8	M	20	93.5	Anomic	0	2.08	2.71	0.63	41	47	6	9.56	2	18
104	60.0	62.4	M	18	79	Anomic	0	2.36	7.33	4.97	45	51	6	1.21	3	19 b
140	75.5	76.3	M	14	91.4	Anomic	0	0.81	1.22	0.41	45	54 a	9	21.95	2	17
320	50.1	50.8	M	16	90.1	Anomic	0	0.66	1.11	0.45	45	48	3	6.67	2	37
1158	72.5	72.9	M	12	96.1	Anomic	0	0.39	0.97	0.58	45	48	3	5.14	2	24
001	58.2	58.5	M	12	93.2	Anomic	0	0.28	1.01	0.73	46	47	1	1.38	4	22
107	36.7	37.5	F	14	93	Anomic	0	0.73	8.11	7.38	47	51	4	0.54	5	15 b
304	59.4	61.3	F	18	92.4	Anomic	0	1.92	2.61	0.69	48	52	4	5.79	2	27
003	76.4	77.0	M	16	92.4	Anomic	0	0.61	3.28	2.67	49	54 a	5	1.87	3	33 c
301	63.0	67.1	M	12	86.4	Anomic	0	4.01	6.22	2.22	49	57 a	8	3.60	3	49 b
190	50.8	51.2	M	12	82.2	Anomic	1	0.35	2.04	1.69	50	54 a	4	2.37	3	49 b
307	46.2	50.9	M	16	80.7	Anomic	0	4.64	5.90	1.25	50	49	−1	−0.80	2	53
184	37.1	37.6	M	12	88.4	Anomic	0	0.49	5.36	4.87	51	57 a	6	1.23	3	16

Note. CVA = cerebrovascular accident (stroke); Ed = education; WAB AQ = Western Aphasia Battery Aphasia Quotient; TPO = time postonset; BNT = Boston Naming Test; Tx = treatment; F = female; M = male.

^a Indicates performance within the normal range.

^b Plus group treatment.

^c Group treatment only.

Funding Statement

The work reported in this article was supported, in part, by Grant DC-R01007646 from the National Institute on Deafness and Other Communication Disorders awarded to P. M. Beeson.