20Q: Criteria for Diagnosing Apraxia of Speech (AOS) in Adults

From the Desk of Ann Kummer

Apraxia of Speech (AOS) is a neurologic speech disorder that causes an impairment in the ability to program and coordinate sensorimotor commands, which are necessary for the production of normal speech. In adults, AOS is caused by acquired neurological conditions, such as stroke, traumatic brain injury (TBI), tumors, or neurodegenerative diseases.  AOS often co-occurs with other types of apraxia (e.g., limb apraxia, apraxia of gait, oral apraxia, and apraxia of swallowing). It is also commonly associated with both dysarthria and aphasia.

In this 20Q article, Katarina L. Haley, Ph.D., CCC-SLP will explain how AOS differs from dysarthria and also how it differs from aphasia with phonemic paraphasia. She will discuss the criteria that is used today to diagnose Apraxia of Speech (AOS) in adults after stroke and other adult-onset neurologic conditions. Finally, she will identify specific strategies that are useful in making an appropriate differential diagnosis.

Here is a little about Dr. Haley and her research. Dr. Haley grew up in Sweden and graduated from the Karolinska Institute in Stockholm. She worked as a certified speech-language pathologist, or “logoped,” in Sweden until 1990. She then moved to the United States to attend Vanderbilt University, where she earned both a Master’s degree and PhD. Dr. Haley is currently a Professor in the Division of Speech and Hearing Sciences at the University of North Carolina, Chapel Hill School of Medicine. She is also a co-director of the UNC Center for Aphasia and Related Disorders. Dr. Haley’s research is focused on the diagnosis, assessment, and treatment of acquired apraxia of speech and on collaborate intervention models for adults living with aphasia.

In this 20Q article, Dr. Haley will help us to learn more about the diagnosis of AOS in adults based on her experience and research. I think you will find this article to be both interesting and informative.

Now…read on, learn, and enjoy!

Ann W. Kummer, PhD, CCC-SLP, FASHA, 2017 ASHA Honors
Contributing Editor 

Browse the complete collection of 20Q with Ann Kummer CEU articles at www.speechpathology.com/20Q

20Q: Criteria for Diagnosing Apraxia of Speech (AOS) in Adults

Katarina Haley

1. I know that Apraxia of Speech (AOS) is a problem with motor programming. Is there more to the definition, or is that what I should be looking for when making a diagnosis?

You are correct, but this definition will only help a bit. AOS is defined as a disorder of the planning or programming of speech movements. Because we assign the breakdown to a stage that is conceptually intermediate between language processing and motor execution, we have also defined what AOS is not. It is not a dysarthria, because muscle weakness, paralysis, or involuntary movements do not explain the difficulties. It is not an aphasia, because limitations in word retrieval, grammatical formulation, auditory comprehension, or phonological assembly do not explain the difficulties either. To complicate matters, people with AOS almost always have aphasia and often dysarthria as well. It is uncommon to have only AOS.  

The AOS definition does not assume a specific etiology other than a neurologic condition that affects the language-dominant cerebral hemisphere (usually the left). It is most often caused by stroke, but can also result from conditions like progressive disease, tumor, and trauma. Knowing about other signs and symptoms, the onset and evolution of the problems, and neuroimaging results will help you make a correct diagnosis.

Knowing that AOS is defined as a speech planning/programming disorder does not do much more for you diagnostically, because this definition is abstract and underspecified. We do not yet understand how brain networks actually program speech movements or why and how people with AOS have the type of difficulties they do. In other words, we don’t have a way to translate our conceptual definition to a behavioral definition. Yet, it is the speech behaviors that speech-language pathologists must rely on when they make an AOS diagnosis.

2. I worry that I may be diagnosing AOS inaccurately. It feels like there are mixed messages about what criteria we should be using. Why is it so confusing?

Many speech-language pathologists feel the same way. Judgments about AOS are very complex, yet our field has been in the habit of assuming that experienced clinicians intuitively and reliably recognize AOS whenever they encounter it. Consider for example the contrast in many aphasia batteries where specific and varied information is gathered about the language profile, but the AOS diagnosis is documented as a simple checkbox. We need to use a more systematic diagnostic process.

There are many different lists with criteria to keep in mind and there are conflicting recommendations about which of those criteria we are supposed to use. These lists with criteria have been assembled from various sources. They are often based on clinical opinion or on studies that have used small sample sizes and a narrow focus. Most studies have had adequate resources to focus on one or two speech features, but very rarely have they been able to consider a more complete syndrome profile. This is one of the reasons we have learned a lot about features that occur in the speech output and much less about what features actually help with diagnosis.

3. So it isn’t enough that a feature is present and instead we must know what features differentiate among disorders? Has this research been done?

That is correct. If the purpose is differential diagnosis, it is essential to focus on features that have been demonstrated to differentiate. Research is progressing to help clinicians, but there is still a lot of work to do. As I mentioned, one of the problems is using very small sample sizes. Small samples always make it difficult to know whether findings are due to real phenomena or chance.

Unfortunately, an even bigger problem with AOS is that it is not possible to conduct normal analyses of diagnostic specificity and sensitivity. Researchers run into a major obstacle as soon as they try to form comparison groups: There is simply no way to know for sure if someone has AOS! It is not possible to use our conceptual definition to determine whether motor programming itself is disrupted. There is also no single test or performance we can use as a reference standard for the behavioral syndrome that is AOS. Instead, the disorder is diagnosed by subjective judgement. This problem has led to confusing and sometimes contradictory diagnostic recommendations.

4. Are you saying that researchers are also confused about how to diagnose AOS? 

Yes, in a way. Individual researchers may feel confident about their gestalt impressions, but they do not all use the same internal reference points. Therefore, diagnostic reliability can be limited even among highly experienced diagnosticians (Haley et al., 2012). Designing studies around this subjectivity can involve less-than-ideal compromises. I’ll share three common research design scenarios that have contributed to confusing conclusions in our literature.

The first scenario is when people who have difficulties with speech are assigned to an AOS group and people who mostly do not have difficulties with speech are assigned to a control group. In these studies, it may seem that speech features that differ between groups are diagnostically important, when in reality they simply differentiate between those who have difficulties and those who mostly do not.

A second scenario is when researchers assign some people to an AOS group based on meeting a set of diagnostic criteria and others to a control group based on not meeting them. It is both unsurprising and unimportant if results show that the groups differ on the same speech features that were used to form the groups in the first place. Results may be interpreted as validating diagnostic criteria when they actually only validate the grouping strategy.

A third scenario is that comparison groups are formed based on clinical judgment, but the grouping criteria are undisclosed or incorrect. If certain speech features differ between an AOS group and a comparison group, we cannot verify diagnostic accuracy and consequently don’t know whether those differences have anything to do with how AOS is currently diagnosed. This has been a particular problem as diagnostic criteria have changed over time.

As you can see, doing research in this area can be tricky. It is important to be disciplined and to seek converging evidence from different research designs and independent research groups. Researchers must be careful to control for differences in severity, avoid circularity of reasoning, and document participants’ speech profiles in detail. With those assurances in place, we can grow our understanding slowly and surely.

5. That explains a lot. Are there other reasons it is difficult to diagnose AOS?

Certainly. First and foremost, it is a complex disorder. AOS varies vastly in severity and quality and its features overlap with both dysarthria and aphasia. Let’s focus on three of the main challenges. The first is that even people who fit the profile of AOS can sound extremely different from one another. Some cannot produce any intelligible speech, whereas others, at first glance, seem to have no difficulties at all. Many have a nonfluent coexisting aphasia, whereas others use long sentences, and a select few may show no evidence of aphasia. Because of these vast differences, you have to be systematic in your assessment and know what you are looking for.

A related challenge is that some cases have a speech profile that is prototypically AOS, but many others are less clear-cut. Diagnosing the former is obviously easier. Research from my own lab suggests that a diagnosis should not be forced in the latter case and that there is far more to the story than a simple presence or absence of AOS. It should be okay to say that the speech profile is uncertain.

Finally, let’s return to the problem that behavior definitions are lacking. This means that speech-language pathologists have no choice but to make rather subjective interpretations if they want to make a diagnosis. If given the opportunity to participate in perceptual team training or other forms of calibration, they can adjust their interpretations to be similar to each other or to a standard. Most of us don’t have such opportunities. As a simple example, a clinician may have learned that she is supposed to listen for abnormal prosody, but not know what aspect of the prosody is most relevant, how much deviation would warrant a diagnosis, or how to weight an abnormal prosodic feature she observes relative to other attributes that are also present. Most of us would benefit from far more perceptual training than we received in graduate school and continuing education courses.

6. What are the current recommendations as far as diagnosing AOS? 

The recommended criteria depend on what you attempt to differentiate. Let’s assume that we want to differentiate people who have AOS (the vast majority of whom also have aphasia) from people who have aphasia without AOS. To make it real, let’s exclude people who have aphasia and make no speech sound errors at all (it is easy to determine that they do not have AOS!).

People with aphasia and people with AOS make errors that sound phonemic to our ears. The first thing to understand is that even though we hear these errors as phonemic, they can be caused by a variety of problems, including a motor planning or programming impairment, a linguistic-phonologic impairment, both, or even a problem with motor execution. The term we use—phonemic paraphasia—refers to a behavior that occurs in both AOS and aphasia.

It does not make sense to differentiate the behavior of phonemic paraphasia from the multidimensional disorder of AOS. It would be like differentiating coughing from the flu. Instead, the diagnostician’s task is to differentiate AOS (which almost always co-occurs with aphasia) from cases of aphasia where phonemic paraphasia is present but due exclusively to linguistic-phonologic etiology. From now on, I will refer to this latter presentation as “aphasia with phonemic paraphasia (APP)”. Defined this way, AOS and APP are both disorders. They are also mutually exclusive. If we follow agreed-upon diagnostic criteria, we can confidently differentiate between the two based on speech qualities we hear. We may not be able to diagnose each case, but we can be confident about the diagnoses we do make.

7. So what are the current diagnostic criteria for differentiating between AOS and APP?

The first agreed-upon criterion is that people with AOS produce a lot more distortion errors and distorted substitution errors than people with APP. The other differentiating criteria are that people with AOS use slow speaking rate, lengthening of pauses, consonants, and vowels, and often equalized stress, whereas people with APP do not (Duffy, 2020; McNeil et al., 2009; Strand, 2014).

8. I’m going to need more detail. First, tell me about the distortion and distorted substitution errors.  What are they?

Distortion errors are subtle phonetic errors you hear beyond the more obvious phonemic ones. They may sound like slurred speech or speech that is not part of the speaker’s native phonology (Wilson et al., 2010). A practical way to define distortion errors is that they are what you would transcribe with diacritic marks if you were using narrow phonetic transcription and heard a deviation that wasn’t indicative of coarticulation, dialect, or other normal variation. Studies from different laboratories and different research designs show that distortion errors occur more often in speech produced by people with AOS than in people with APP (Bislick et al., 2017; Haley et al., 2017). We still have to work out the cutoff values from normative studies, but hearing a lot of sound distortion errors is an agreed-upon sign of AOS.

Remember that phonemic errors, such as substitutions, occur in both AOS and APP, so even though they certainly characterize these disorders, it is one of the many features that don’t differentiate between them. Distorted substitution errors, on the other hand, refer to phonemic errors and distortion errors that are produced on the same segments and they occur more often in AOS than in APP. Of course they are only a small subset of all the phonemic and distortion errors that people with AOS produce and they can be difficult to catch.

9. Can you give some examples of distortion and distorted substitution errors?

One common type of distortion error is an ambiguous (i.e. blurred) distinction between voiced and voiceless consonants. For example, when saying the word “boss,” the first consonant may sound like something in between a /b/ and a /p/. When you listen to the production several times, you may find that your perceptions shift. Sometimes you hear /b/ and sometimes you hear /p/; perhaps you can’t decide which it is. Other common examples are when one of the segments is unusually long (e.g. a prolonged /s/) or the constriction is unusually open (e.g. when the /b/ sounds almost like a fricative).

An example of a distorted substitution would be saying something that sounded like “toss” instead of “boss” (substitution of the first consonant) and at the same time using ambiguous voicing, so it wasn’t actually clear whether the first consonant was a /t/ or a /d/ (distortion of the substituted consonant). Distorted substitutions can also be distorted additions. In our example, “boss” could be produced as “bossa” with a very prolonged final vowel. 

10. And then it seems you mentioned several other criteria in one breath…. slow speaking rate, lengthening of pauses, consonants, and vowels, and equalized stress? What are those?

Yes, I like to mention them together, because they are closely interrelated. We just talked about distortion errors and that one example is lengthening of a consonant or vowel segment. Well, lengthening also affects the rhythm and rate of speech. Besides lengthening of consonants and vowels, people with AOS also insert and lengthen pauses within words. They may produce “boss” as “bo…ss” with a long pause between the vowel and final consonant. Together all those lengthening behaviors result in slower than normal speech.  The clearest way to see the effects is when saying words that include three or more syllables. In fact, if you calculate the mean length of syllables in multisyllabic words (duration of the word divided by number of syllables produced), you have a very good index of whether the rate is abnormally slow (more than 300-350 ms). We call this measure the Word Syllable Duration (WSD) and have found it to be very helpful for diagnostic decision-making (Haley et al., 2012; Haley & Jacks, 2019). It is also a good outcome measures for documenting change over time.   

Sometimes the pauses and sound prolongations make it sound like the speaker produces words one syllable at a time. This phenomenon, syllable segmentation, is particularly striking when there is a pause between each consecutive syllable. Prolongations can also equalize stress differences. Take the word “catastrophe” for example. The vowel in the first syllable is normally unstressed and very short, whereas the vowel in the second syllable is stressed and long. People with AOS often lengthen unstressed syllables with the effect that all vowels and syllables are of approximately the same length. This is one example of how prosody is impaired in AOS. Prosodic difficulties are restricted to these timing problems in AOS, and variations in voice frequency (intonation) are relatively spared.

11. I feel like there are other criteria we haven’t talked about. What about error variability for example? It’s often mentioned.

Error variability is one of those features that are common, but don’t help you differentiate between AOS and APP (Haley et al, 2020; Staiger et al., 2012). Except when the speech targets are too easy or too difficult, the way people with either disorder produce words tends to vary from attempt to attempt. For example, if you ask a person with moderately severe AOS to say the word catastrophe three times, they may say something that sounds like “ca..sta..fafi”, “casasafi,” “cast..rafri.”

The nature of the variability also means that any given phoneme can be altered in a number of ways. Sometimes /b/ may be produced as /g/, sometimes as /sp/, and so forth. On the other hand, individual people with AOS tend to favor certain sounds and sound combinations, which means that you can identify patterns to address in treatment (Wambaugh et al., 1998). People also tend to produce the same proportions of phonemic and distortion errors over time, so documentation is straightforward and reliable (percent accuracy, intelligibility, etc; Haley & Martin, 2011; Mauszycki et al., 2010).

12. What about groping? That is also something I have seen used as a diagnostic criterion.

The word groping refers to a type of searching behavior, where the speaker self-corrects and attempts different ways to say a word, syllable, or sound. Searching behaviors also occurs in APP. We use different words for the phenomenon in people with aphasia (e.g. conduite d'approche), but it has so far not been shown that the actual behavior is different.

13. What about increasing errors with longer utterances?

It is true that people with AOS have more difficulties with longer words than with shorter words, but so do people with APP and people with dysarthria.

14. What about producing more errors on vowels than on consonants?

Again, there is no research evidence that this consonant-over-vowel pattern differentiates between AOS and APP. It’s also true that the extent to which you hear errors on vowels is highly dependent on the task. Speech-language pathologists, like other proficient speakers of a native language, tolerate more variations in vowels before they detect an error, especially when they know the target word, as we so often do in our assessment tasks (Haley et al., 2001). For similar reasons, we are predisposed to detect errors at the beginning of words and syllables rather than at the end.

15. Okay, so that is it? I just feel like I’ve seen lists with a lot more criteria listed.

Yes, that is it. These are the diagnostic criteria for which research supports a distinction between AOS and APP. Again, there are many other features of the speech output that may be striking, yet do no not help with differential diagnosis as far as we know. We should expect that diagnostic criteria will be adjusted and expanded as we continue to learn more about AOS and design studies that compare appropriate diagnostic groups.

Also, so far, we have only discussed the difference between AOS and APP.

16. That’s right! We forgot about dysarthria. How do you differentiate between AOS and dysarthria?

The question of whether a person has dysarthria or AOS comes up more often with progressive disease than with left hemisphere stroke and it can have very important implications for medical diagnosis and prognosis.

The following are six basic observations that would suggest AOS rather than dysarthria (Duffy, 2020; McNeil et al., 2009): 1) Automatic speech, like counting and saying the days of the week, is distinctly more normal than regular speech; 2) Speech difficulties are restricted to articulation and prosody with no evidence of abnormal phonation or hypernasality; 3) The person has considerable difficulties repeating complex and meaningless syllable sequences like “puh-tuh-kuh”; 4) An oral mech exam shows no evidence of muscle weakness; 5) Distortion types are diverse and variable (voicing ambiguity, lengthening, tongue fronting, constriction weakening, resonance ambiguity, etc) rather than uniform and consistent; 6) Sound errors are often corrected, sometimes with repeated attempts.

It is far more difficult to diagnose dysarthria when it coexists with AOS, because you are observing a combination of overlapping and similar features. We do think it is quite common, even for people with stroke AOS to have dysarthria. Presumably, both disorders contribute to the perception that sounds are distorted and speaking rate is slow.

17. Now that we have discussed the first steps to differential diagnosis (knowing what you are differentiating and knowing the diagnostic criteria), let’s talk about the next step. Is there a test I should be giving or some other way to collect an informative speech sample?

The first step is to ask patients to explain their speech problems and how they are affected by them. For people who have recently acquired aphasia, AOS, or dysarthria the evaluation session should connect their personal experiences to more specific information you can provide about their diagnosis and what to expect.

People who have lived with the disorder for some time can often tell you what helps, what hinders, and what they have tried as far as strategies and treatments. I continue to be impressed with the insight many people have if you give them the opportunity to explain what their problems feel like.

18. I appreciate that you mention this interaction as a first step. It is not as if speech-language pathologists make the AOS diagnosis all alone in their offices. We have to process a lot of information while simultaneously interacting meaningfully with our patients. What is the next step?

The next step is to listen to the quality of a connected speech sample. Some of this sample is based on initial conversation. A supplemental elicited monologue (e.g. picture description) sometimes helps the clinician focus on the speech quality more specifically. Because most people for whom you would consider AOS have coexisting aphasia, any number of language problems will make their connected speech less fluent and thorough, resulting—in particular—in hesitations, pauses, slow speaking rate, and sparse output. Additionally, when people construct their own narrative, they may use avoidance strategies to communicate or otherwise complete the assessment task as effectively as possible. For these reasons, connected speech samples can both overestimate and underestimate the speech production difficulties people experience.

The third step is to conduct a word repetition task that is more extensive than what is included in most aphasia batteries. This task almost always gives you the missing information you need for your evaluation. The Apraxia Battery for Adults -2 (ABA-2; Dabul, 2002) includes targets with a range of complexity to help reveal pertinent articulation and prosody challenges. Other motor speech examination protocols are equally suitable. You can find copies in textbooks and research articles (Duffy, 2020; Freed et al, 2020; Haley et al, 2017).

There are presently no comprehensive diagnostic assessments for AOS. Ideally, future tools will allow the clinician to determine how phonetic, lexical, modality, and strategic factors influence speech patterns. Such is the current focus of work in our research lab at the UNC Center for Aphasia and Related Disorders (aphasia.und.edu), so hopefully we can share resources with you in the near future.

19. Earlier you said that knowing about neuroimaging can help speech-language pathologists make a correct diagnosis. Do we know what part of the brain is responsible for AOS?

Only to a limited extent. We know that a large neural network is responsible for speech planning and programming. It is distributed across several cortical and subcortical regions in the brain and different links in this network can disrupt speech production. There is no single region that is responsible for AOS. However, the left posterior inferior frontal cortex (Broca’s area), supplementary motor area, and lateral premotor cortex appear to play central roles (Fridriksson et al., 2018; Josephs et al., 2012; Richardson et al., 2012). In contrast, if the anatomical and functional involvement is only in parietal and temporal cortex you would expect APP or just aphasia without speech production difficulties.

Overlapping language and speech production networks mean that aphasia classifications can also be helpful. For example, AOS is common in Broca’s aphasia, whereas APP is common in Wernicke’s aphasia.

20. Okay, so in summary, I should understand the difference between features that are present in the speech and features that help with differential diagnosis, make sure I administer a motor speech examination, and look for converging evidence. What should I do if I still feel uncomfortable deciding whether my patient has AOS?

Actually, my recommendation is the same whether you feel comfortable or uncomfortable: Document the extent to which you observe features in the patients’ speech that are consistent with current diagnostic criteria. Measure when you can;  then analyze the profile you have documented. You should be able to use your observations as a rationale for the diagnosis.

If you cannot justify the diagnosis or results are inconclusive, it is best to state that the differentiation is uncertain. At least that is what we do in my lab, where our research objective is to better understand relationships among AOS, APP, and dysarthria.