Assessing Swedish Genetic Counselling Outcome Measures for Autism and General Use: Rasch Findings Highlight the Need for Improved Measures

This study presents the first Swedish autism-specific adaptation of the Genetic Counselling Outcome Scale and, through Rasch analysis, reveals that while the modified version offers some usable subscales, both it and the general scale suffer from significant psychometric limitations, suggesting a need for more substantially adapted tools to accurately measure genetic counselling outcomes in autistic populations.

The Gist

The Big Picture: Trying to Measure "Feeling Empowered"

Imagine you are a genetic counselor. You sit down with a family who just found out their child has autism. You explain the genetics, the risks, and the future. Your goal is to help them feel empowered—like they understand the situation, can make good decisions, and aren't overwhelmed by fear.

But how do you know if you actually helped them? You can't just ask, "Did I do a good job?" because people might say "yes" just to be polite. Instead, researchers use a questionnaire (a survey) called the GCOS-24. Think of this survey as a thermometer designed to measure the "temperature" of a patient's empowerment.

This study asked a simple but crucial question: Is this thermometer actually working correctly, especially when used on autistic families?

The Problem: The "Swiss Army Knife" That Was Too Clunky

The researchers took the standard Swedish version of this survey and tweaked it specifically for autistic families (changing words like "medical condition" to "neurodevelopmental condition" so it didn't feel too clinical). They called this the Swedish Autism mGCOS-24.

They then used a very strict, mathematical way of checking the survey called Rasch Analysis. If you imagine the survey as a ruler, Rasch analysis checks if the markings on the ruler are actually evenly spaced. Does the distance between "1 inch" and "2 inches" feel the same as "2 inches" to "3 inches"?

The Findings:
The researchers discovered that the ruler was broken in a few ways:

1. It was trying to measure too many things at once (Multidimensionality):

   • The Analogy: Imagine you have a scale in your kitchen. You put an apple on it, and it tells you the weight. But then you put a heavy book on it, and the scale starts shouting out the color of the book and the temperature of the room.
   • The Reality: The survey wasn't just measuring "empowerment." It was accidentally measuring "knowledge," "hope," "coping skills," and "mental control" all mixed together. Because these are different things, the total score was confusing. It was like trying to add up your height, your weight, and your shoe size to get one "health number." It doesn't make sense.

2. The Steps were confusing (Disordered Thresholds):

   • The Analogy: Imagine a staircase where the third step is actually lower than the second step. If you try to walk up it, you stumble.
   • The Reality: The survey had options like "Strongly Disagree," "Disagree," "Neither," "Agree," "Strongly Agree." The math showed that people didn't see these steps as a smooth ladder. Sometimes, "Neither Agree nor Disagree" felt the same as "Disagree," or "Slightly Agree" felt the same as "Agree." The steps were jumbled, making the measurement inaccurate.

3. It didn't work the same for everyone (Invariance Issues):

   • The Analogy: Imagine a ruler that stretches when you use it on a hot day and shrinks when it's cold. If you measure a table at noon and again at night, you get different numbers, even though the table didn't change.
   • The Reality: The survey gave different results depending on the person's age or gender. A 20-year-old and a 50-year-old might answer the same question differently not because their empowerment is different, but because the question itself is interpreted differently by different age groups.

The Solution: Breaking the Big Ruler into Smaller Rulers

Since the big, 24-question ruler was broken, the researchers tried a different approach. They chopped the survey into smaller, specific pieces (sub-scales).

They created four smaller rulers:

1. Knowledge: How much do you understand?
2. Hope: Do you feel optimistic?
3. Coping: Can you handle the stress?
4. Cognitive Control: Can you make clear decisions?

The Result:
When they tested these smaller rulers, they worked much better! They were straight, the steps were even, and they measured what they were supposed to measure.

However, there was a catch:
These new, smaller rulers worked great for the autistic group, but they didn't work when applied to the general group (people with cancer or other conditions).

• The Analogy: It's like designing a pair of custom boots that fit a specific person's foot perfectly. If you try to force those same boots onto someone with a completely different foot shape, they won't fit. The "autism-specific" questions just didn't make sense to the general population in the same way.

The Takeaway: Why This Matters

The authors are saying: "Stop using the big, messy survey to get a single score."

1. Don't trust the "Total Score": If a clinic uses the old survey and gets a score of "50 out of 100," that number is misleading. It's a mix of hope, knowledge, and fear all rolled into one.
2. Look at the details: Clinicians should look at the specific sub-scores (e.g., "This family has high knowledge but low hope"). This gives a much clearer picture of what the family actually needs.
3. We need better tools: The current tools are like a blunt instrument. We need to build new, sharper tools specifically designed for autistic families, rather than trying to force a "one-size-fits-all" survey to work.

In a Nutshell

The researchers took a popular survey used to measure how empowered families feel after genetic counseling. They found that the survey was broken—it was measuring too many things at once and confusing the steps. By breaking it into smaller, specific questions, they fixed the problem for autistic families, but they realized that a "one-size-fits-all" survey might never truly work for everyone. We need better, more specific tools to truly understand how genetic counseling helps people.

Technical Summary

1. Problem Statement

Genetic counseling is a critical intervention for families affected by autism, yet reliable Patient-Reported Outcome Measures (PROMs) to evaluate its effectiveness are lacking. The Genetic Counselling Outcome Scale (GCOS-24) is the standard instrument for measuring patient empowerment in genetic counseling. However:

• Contextual Gap: While a Swedish version of the GCOS-24 exists, no autism-specific version has been developed or validated.
• Methodological Limitations: Previous validation studies of the GCOS-24 (including the Swedish version) relied heavily on Classical Test Theory (CTT) and factor analysis. These methods often fail to detect specific item-level dysfunctions, local item dependencies, and violations of unidimensionality.
• Generalizability Issues: It is unclear whether the construct of "empowerment" functions identically across different clinical contexts (e.g., general medical conditions vs. autism) or if a single scale can validly measure outcomes across diverse populations.

2. Methodology

The study employed Rasch Measurement Theory (RMT), a modern psychometric framework, to evaluate two instruments:

1. Swedish Autism mGCOS-24: A newly adapted version of the GCOS-24 tailored for autistic individuals and their families.
2. Swedish General GCOS-24: The existing general version used for comparison.

Data Collection:

• Autism Cohort ($N=171$): Recruited from a clinical trial (KONTAKT) and patient organizations. Participants were aged 15–62 (median 28), predominantly female (77%), and 100% diagnosed with autism.
• General Cohort ($N=383$): Reused data from a previous validation study involving individuals scheduled for genetic counseling (mostly cancer patients, median age 47).

Analytical Approach:

• Preprocessing: Items were inverted where necessary, and response categories were shifted to 0-indexing.
• Rasch Diagnostics: The study assessed:
  • Unidimensionality: Using Principal Component Analysis (PCA) of residuals and eigenvalues against simulation-based critical values (a more rigorous standard than rule-of-thumb thresholds).
  • Local Independence: Examining residual correlations between item pairs.
  • Response Category Functioning: Checking for disordered thresholds.
  • Invariance: Testing Differential Item Functioning (DIF) across age and sex using Andersen's Likelihood Ratio (LR) test.
  • Reliability: Measured via Cronbach's alpha, Person Separation Index (PSI), and Relative Measurement Uncertainty (WLE-RMU).
• Iterative Restructuring: When the full scale failed RMT assumptions, the researchers iteratively created subscales based on thematic grouping and collapsed response categories to achieve ordered thresholds.

3. Key Contributions

• First Swedish Autism-Specific mGCOS-24: The study successfully translated and culturally adapted the GCOS-24 for the Swedish autistic population, removing medicalized language and incorporating family perspectives.
• First Rasch-Based Evaluation in Swedish: This is the first study to apply RMT to any Swedish version of the GCOS-24 or mGCOS-24, moving beyond traditional CTT metrics.
• Methodological Rigor: The authors utilized simulation-based critical values for eigenvalues and conditional item fit statistics, addressing known limitations in previous Rasch studies that used unconditional fit statistics or rule-of-thumb cutoffs.
• Subscale Development: The study identified that the scale is not unidimensional and proposed a four-subscale structure for the autism context.

4. Key Results

A. Full Scale Performance (Both Cohorts)

• Multidimensionality: Both the Swedish Autism mGCOS-24 and the General GCOS-24 failed the unidimensionality test. The first eigenvalue of residuals significantly exceeded simulation-based thresholds (e.g., 5.08 vs. 2.06 for the autism scale), indicating the scales measure multiple distinct latent constructs rather than a single "empowerment" factor.
• Disordered Thresholds: All items exhibited disordered response thresholds, meaning the 7-point Likert scale did not function as an ordinal progression. Respondents could not distinguish between adjacent categories (e.g., "Slightly agree" vs. "Agree") in a meaningful way.
• Lack of Invariance: The full scales showed significant Differential Item Functioning (DIF) regarding age (and sex for the general scale), meaning item difficulty varied by demographic group, invalidating total score comparisons.

B. Subscale Performance (Swedish Autism mGCOS-24)
By restructuring the scale into four thematic subscales and collapsing response categories, the researchers achieved acceptable psychometric properties:

1. Knowledge
2. Hope
3. Coping
4. Cognitive Control

• Results: These subscales demonstrated unidimensionality (eigenvalues < simulation thresholds), ordered thresholds, local independence, and invariance across age and sex.
• Reliability: Reliability ranged from low (Knowledge, PSI $\approx$ 0.55) to medium/high (Hope, Coping, Cognitive Control, PSI $\approx$ 0.75–0.79).

C. Generalizability Failure

• When the four subscales derived from the autism cohort were applied to the General GCOS-24 data, they failed to maintain unidimensionality and invariance (except for the "Coping" subscale).
• This indicates that the measurement structure is context-specific; items function differently for autistic individuals compared to the general clinical genetics population (largely cancer patients).

5. Significance and Implications

• Rejection of the "Total Score" Approach: The study concludes that using a single total score for the GCOS-24 is psychometrically invalid for both general and autism-specific contexts in Sweden. The construct of empowerment is multidimensional.
• Need for Context-Specific Tools: A "one-size-fits-all" genetic counseling outcome measure may not be feasible. Instruments must be tailored to specific populations (e.g., autism) to ensure measurement invariance.
• Response Format Improvements: The disordered thresholds suggest the current 7-point Likert scale is too granular or confusing. The authors recommend:
  • Collapsing response categories (as done in the subscales).
  • Adding an "Irrelevant" option or allowing item skipping, as the current instruction to select "neither agree nor disagree" for irrelevant items violates ordinal assumptions.
• Future Research Direction: Future development of PROMs for genetic counseling should prioritize RMT validation over CTT. If a total score is required for practical reasons, it should be calculated as a standardized average of subscale scores, interpreted with caution as a composite index rather than a true Rasch measure.

Conclusion:
The study highlights that while the Swedish autism mGCOS-24 can be salvaged into four valid subscales, the broader GCOS-24 instrument requires substantial refinement. The findings underscore the necessity of developing new, or more heavily adapted, tools to accurately capture the nuanced outcomes of genetic counseling in autistic populations.