Error behaviors associated with loss of competency in Alzheimer’s disease

Capacity to Consent to Treatment Instrument.

Patients and controls were administered the CCTI.1 As previously reported,1 the CCTI comprises two clinical vignettes (A, neoplasm; B, cardiac) and tests the capacity to consent under five LSs. After presentation of a vignette, individuals are asked a series of questions that test their capacity to consent under each of five well-established LSs.14,15 The LSs are presented below in order of increasing stringency1:

• LS1 = simply evidencing a treatment choice;

• LS2 = making the reasonable treatment choice (when the alternative is unreasonable) (vignette A only);

• LS3 = appreciating the consequences of a treatment choice;

• LS4 = providing rational reasons for a choice; and

• LS5 = understanding the treatment situation and choices.

Quantitative scoring system for the CCTI.

As previously reported,1 a detailed and well-operationalized quantitative scoring system with high inter-rater reliability was developed for the five CCTI LSs. This quantitative scoring system has demonstrated face and content validity and has successfully discriminated the LS competency performance of normal older controls as well as mild AD and moderate AD patients.1 The CCTI is a reliable and valid instrument for quantitatively assessing capacity to consent.

Qualitative scoring system for the CCTI.

We also developed a system of qualitative scores for identifying behavioral changes and errors reflected in AD patient responses on the CCTI test items. Some error codes were drawn from the Exner special scoring system for the Rorschach personality test8 (deviant verbalization, alogical reasoning, personalization),8 but were modified slightly to better represent verbal behaviors of AD patients. The majority of the error codes were developed based on the investigators’ experience with control and AD patient verbal responses produced during piloting and standardization of the CCTI. As described below, error codes were initially viewed as either benign or more pathognomonic based on this experience.

The qualitative scoring system consists of 16 mutually exclusive error codes conceptually organized into four domains: 1) language dysfunction, 2) executive dysfunction, 3) affective dysfunction, and 4) compensatory responses (table 1 provides specific definitions and actual examples for each of the codes). The individual codes were developed based on behaviors that had first been noted during pilot testing of the CCTI. They were then grouped into domains based on behavioral similarities and the authors’ initial views regarding underlying mechanisms. Codes in the language dysfunction domain identify changes in both expressive and receptive language, and range from relatively benign errors such as circumlocution and tangentiality to more serious language errors such as paraphasias and neologisms (deviant verbalization) and comprehension failures (miscomprehension). Codes in the executive dysfunction domain identify a variety of problems in verbal reasoning, processing, and production. Codes in this domain range from relatively mild factual errors (factual confusion), strained reasoning (alogical reasoning), and intrusion errors (e.g., an individual’s confusion of information between the two CCTI vignettes) to more serious processing breakdowns such as answers completely unresponsive to the question (nonresponsive answers), unintelligible responses (incoherence), loss of detachment from the hypothetical medical problem (i.e., an individual’s belief that the vignette scenario is an actual personal medical problem), and loss of task altogether. The affective dysfunction domain consists of one code for depressive verbalizations (melancholic responses). Codes in the compensatory responses domain identify behaviors that reflect potential compensatory adaptations to the competency task by both normal elderly individuals and cognitively impaired patients. These codes range from old learning adaptations such as personalized responses (personalization) and personal value statements (personal appreciation) to more problematic adaptations such as unwarranted inferences not justified by the vignette context and outright delegation of medical decision-making responsibility to others.

Considerable effort was made to ensure that the codes identified distinct behaviors and were mutually exclusive. Detailed and well-operationalized scoring criteria were developed for each of the codes. (Because of space limitations, the scoring protocol cannot be presented here but may be obtained by contacting the first author). A conservative approach was adopted in assigning codes to verbal responses, with a maximum of one code permitted for each conceptual unit of verbal text.

Reliability of the error code scoring system.

Before the present study, we conducted a reliability study of the error codes using 23 individuals (9 controls and 14 patients with AD). Three raters trained in scoring the error codes and the CCTI LSs achieved acceptable levels of inter-rater reliability using exact agreement as a conservative criterion. Overall percentage agreement was 83.21% for vignette A (n = 322 text observations over the 23 cases), 79.79% for vignette B (n = 322 observations over 23 cases), and 81.50% for the entire protocol (n = 644 observations over 23 cases). All three raters agreed on the assignment of codes in 421 of the 644 observations. Two of the raters agreed on error codes for 155 of the remaining 223 observations, and there was no rater agreement for 68 of the observations. Because the raters were scoring complex behaviors from verbal transcripts, some classification ambiguity was expected, and we were pleased with the relatively high level of agreement obtained.

CCTI administration procedures.

The two CCTI vignettes were administered in counterbalanced order across study participants (n = 93). Each CCTI protocol was audiotaped during administration and subsequently transcribed to a written protocol before review and scoring. The results of the prior reliability study were used to train two new raters who computed LS scores and assigned error codes. All error codes for the current study’s 93 CCTI protocols were reviewed and finalized by a third, master rater (B.M.) who was blinded to diagnosis.

Neuropsychological measures.

As part of a larger study of the CCTI, neuropsychological measures commonly used in dementia evaluation were also administered to study participants. These measures represented cognitive domains linked conceptually and empirically to the capacity to consent to treatment.16,17 These domains and measures were as follows:

Group comparisons on demographic, CCTI LS, and neuropsychological variables.

Control (n = 21) and AD (n = 72) groups were compared on demographic, CCTI LS, and neuropsychological variables using t-tests.

Group comparisons on error codes.

Incidence of error codes was compiled within-group for each LS of each vignette and for all CCTI protocols. Given the nature of the data (frequency information that was not normally distributed), nonparametric tests (Mann-Whitney28) were used. Planned comparisons of error incidence were made between control (n = 21) and AD (n = 72) groups, control and mild AD (n = 47) groups, control and moderate AD (n = 25) groups, and mild AD and moderate AD groups. These contrasts were planned so that the pattern of error behaviors across the course of AD could be examined.

Error code predictors of CCTI LS performance within group.

As previously reported,1,17 quantitative scores on the CCTI LSs were summed across vignettes A and B to create composite LS variables (LS1, LS3–LS5) (LS2 is unique to vignette A). Regression analyses were then run within control and AD groups to identify error types that predicted performance on LSs. Depending on LS score range, either logistic or linear regression was used. Because controls performed perfectly on LS1 and LS2, these LSs were not included in the analysis for controls.

Neuropsychological predictors of error codes in the AD sample.

We identified neurocognitive correlates of selected error codes in the AD sample using univariate Spearman correlation and linear regression. Error codes were selected if they were predictive of LS performance in the competency analysis (above). The neuropsychological measures with the highest univariate correlations with each specified error code were entered into regression equations to identify neurocognitive predictors of that code.

Language dysfunction codes.

Those in the AD group demonstrated more miscomprehension of questions than controls. Compared with controls, patients with moderate AD produced significantly more deviant verbalizations and miscomprehensions. The mild and moderate AD groups did not differ significantly in incidence of language dysfunction errors. There were no group differences in incidence of circumlocution or tangentiality (see table 3).

Executive dysfunction codes.

Compared with controls, patients with AD had significantly more instances of factual confusion, intrusion, incoherence, nonresponsive answer, and loss of task. Patients with mild AD demonstrated more factual confusions, intrusions, nonresponsive answers, and losses of task than controls. Patients with moderate AD produced more intrusions, incoherent responses, nonresponsive answers, and losses of task than controls. Compared with mild AD patients, moderate AD patients also demonstrated more incoherence, loss of detachment, nonresponsive answers, and loss of task. Nonresponsive answer and loss of task were the only error behaviors that discriminated all groups. There were few instances of alogical reasoning and no significant group differences on this error code (see table 3).

Affective dysfunction code.

Moderate AD patients made more melancholic responses than controls or mild AD patients (see table 3).

Compensatory response codes.

The AD group produced more delegation responses than controls, with moderate AD patients producing more delegation responses than either controls or mild AD patients. No group differences were found for personal appreciation, personalization, or unwarranted inference.