Kevin P. Weinfurt

Background: Concerns about measurement validation are often expressed as imperatives to use a “valid measure”. However, validity is not a characteristic of a measure. Instead, validation and validity refer to score interpretation within a context of use. This is important because responses to health measures can be different in different contexts, influencing equitable consequences of measurement. In this paper, we aim to (1) outline the theory of the argument-based approach to validity, and (2) discuss assumptions and evidence in relation to equitable consequences of measurement. Methods: The argument-based approach to validity asks us to first state how scores will be interpreted and used in context. Assumptions underpinning this statement guide validation planning. Existing and new evidence need to be examined and evaluated in relation to the assumptions and concept of interest, leading to a reasoned evidence-based argument about the degree to which score interpretation in a context of use is valid, with consideration of potential threats to validity and measurement consequences. Results: Key assumptions are described, including why evidence is needed, what evidence tells us, and the importance of assumptions in relation to the equitable consequences of measurement. Conclusion: The argument-based approach to validity shifts the focus of validation to a score’s interpretation and use in a context, in relation to the concept of interest. A validity argument is built from evidence about the plausibility of the score interpretation in the context of use with consideration of the degree to which measurement consequences will lead to the intended beneficial health consequences and not perpetuate existing inequities in health.

This commentary consists of edited remarks made by the author upon receipt of the 2025 International Society for Quality of Life Research President’s Award. The field of health measurement/quality-of-life research is viewed as a house in which we have grown up and now live. We sometimes need to rework our foundation to maintain our field’s structural integrity —that is, the shared intellectual foundations that support our thinking in a changing world. To complement structural integrity, we need moral integrity—a shared commitment to working in ways that strengthen our science and our humanity. Monitoring and sustaining these two aspects of integrity are the responsibility of everyone.

Human Immunodeficiency Virus (HIV) criminalization laws may discourage engagement in care. American Men's Internet Survey participants were randomized to disclosure, in which automatic public health reporting of antiretroviral resistance testing (ARVRT) was explained upfront, or non-disclosure, in which it was revealed later. Among 4,750 participants, 64.4% were unsure whether their state had an HIV criminalization law. ARVRT willingness was greater with non-disclosure (OR = 1.96, 95% CI: 1.41-2.71) and most pronounced among those believing no law existed (OR = 4.00, 95% CI: 1.10-14.54), with an interaction between legal perception and disclosure condition (χ² (2) = 7.14, p = 0.028).

Background: Terheyden et al. recently described a compelling vision for large language model-enabled patient-reported outcome measures (LLM-PROMs). Main text: We support Terheyden et al.’s vision and offer complementary observations about the potential for generative artificial intelligence (GenAI) in assessing patient-centered outcomes. GenAI has the potential to improve the quality and efficiency of developing traditional PROMs and collecting patient experience data. Traditional PROMs rely on standardized questions and responses, which may introduce ambiguity about the health concept being assessed. Yet, interviewers who are trained in the meaning of the concepts can tailor questions to the respondent’s experience and conversation style and have a back-and-forth clarification of meaning to ensure that both the interviewer’s and respondent’s meanings are aligned. The shortcoming of this approach is that it cannot be done at scale with human interviewers. However, trained GenAI interviewers could make such an assessment a reality for large samples of patients. The technology is already available to train GenAI interviewers in interview technique, the intent of each item, and a consistent approach toward coding the respondent’s answer based on the conversation. Conclusion: The health outcomes research field should actively inquire into what patient experience data can be collected via GenAI and rigorously evaluate the quality of the assessments obtained.

Introduction: Understanding how different symptoms co-occur and are correlated may provide insights into the pathophysiology of disease. The lack of group-to-individual generalizability of co-occurrence of symptoms was recently demonstrated by comparing intra-individual and inter-individual correlations in several psychological studies. Here, we investigate this phenomenon for lower urinary tract symptoms (LUTS). Methods: We analyzed data collected in the Symptoms of Lower Urinary Tract Dysfunction Research Network Recall Study. Participants responded to questions about their urinary symptoms for 25 consecutive days. These questions queried urologic symptoms including storage (urinary urgency, frequency, nocturia, and urinary incontinence), voiding (slow/weak stream), and post-micturition (incomplete emptying and post-micturition dribble) symptoms. We calculated Pearson correlation coefficients and cosine similarity measures and compared distributions of intra-individual and inter-individual (cohort) metrics. Results: Among 234 participants, distributions of intra-individual measures were 10-fold wider than those of inter-individual correlations. There are pairs of questions with distributions of correlations and cosine similarities containing individuals with extreme positive (>0.8) and extreme negative values (<–0.8). There are groups of participants with strong positive and negative correlations of urinary frequency and nocturia, urinary incontinence and weak flow, as well as strong negative and positive correlations of urinary frequency and dribbling. Information on these extreme groups is averaged out and lost in the inter-individual correlations. Conclusions: Lack of group-to-individual generalizability previously shown for psychological symptoms is confirmed for LUTS. Wealth of information on the co-occurrence and co-evolution of LUTS in the intra-individual correlations and cosine similarities corroborates heterogeneity of LUTS and can be useful for deep phenotyping and for identifying personalized treatments of LUTS.

Introduction: Biostatisticians increasingly use large language models (LLMs) to enhance efficiency, yet practical guidance on responsible integration is limited. This study explores current LLM usage, challenges, and training needs to support biostatisticians. Methods: A cross-sectional survey was conducted across three biostatistics units at two academic medical centers. The survey assessed LLM usage across three key professional activities: communication and leadership, clinical and domain knowledge, and quantitative expertise. Responses were analyzed using descriptive statistics, while free-text responses underwent thematic analysis. Results: Of 208 eligible biostatisticians (162 staff and 46 faculty), 69 (33.2%) responded. Among them, 44 (63.8%) reported using LLMs; of the 43 who answered the frequency question, 20 (46.5%) used them daily and 16 (37.2%) weekly. LLMs improved productivity in coding, writing, and literature review; however, 29 of 41 respondents (70.7%) reported significant errors, including incorrect code, statistical misinterpretations, and hallucinated functions. Key verification strategies included expertise, external validation, debugging, and manual inspection. Among 58 respondents providing training feedback, 44 (75.9%) requested case studies, 40 (69.0%) sought interactive tutorials, and 37 (63.8%) desired structured training. Conclusions: LLM usage is notable among respondents at two academic medical centers, though response patterns likely reflect early adopters. While LLMs enhance productivity, challenges like errors and reliability concerns highlight the need for verification strategies and systematic validation. The strong interest in training underscores the need for structured guidance. As an initial step, we propose eight core principles for responsible LLM integration, offering a preliminary framework for structured usage, validation, and ethical considerations.

Background: Molecular HIV surveillance (MHS) can be used to help identify and respond to emerging clusters of rapidly spreading HIV transmissions, a practice known as cluster detection and response (CDR). In the United States, MHS relies on HIV gene sequences obtained from routine clinical antiretroviral resistance testing (ARVRT). By law, ARVRT results are reported to public health agencies for MHS and individuals are not asked for their specific consent to do so. This practice has raised ethical concerns, including the lack of consent for, and transparency surrounding, public health uses of these clinical data. Such concerns have spurred debate and could have a chilling effect on the willingness of people living with HIV to agree to ARVRT when recommended clinically and jeopardize the utility of MHS-informed HIV prevention efforts. In response to the lack of routine disclosure of use of ARVRT results for MHS, in 2022, the Presidential Advisory Council on HIV/AIDS (PACHA) issued a resolution calling on the US Centers for Disease Control to “require that providers explain MHS/CDR and the laboratory test results that are collected and used in these surveillance activities to their patients.” Objective: This study aimed to examine the effect of clinician disclosure of the public health uses of ARVRT results for MHS versus clinician nondisclosure on patient willingness to undergo recommended ARVRT. Methods: We conducted a randomized survey experiment examining the effect of clinician disclosure of the public health uses of ARVRT results for MHS versus clinician nondisclosure (the current standard of care) and subsequent discovery of such uses through a “trusted media source” on patient willingness to undergo recommended ARVRT. Study participants were respondents to 1 of 2 national web-based surveys conducted annually in the United States: the American Men’s Internet Survey (AMIS) and the Transgender Women’s Internet Survey and Testing (TWIST). Results: Overall, 4348 AMIS participants (n=2151 disclosure; n=2197 nondisclosure) and 3314 TWIST participants (n=1670 disclosure; n=1644 nondisclosure) completed survey items regarding the randomly assigned vignettes. The majority were willing to undergo ARVRT regardless of which vignette they saw (1670/2151, 82.7% [AMIS] and 1326/1670, 80.8% [TWIST] in the disclosure group; and 1399/2197, 68% [AMIS] and 1101/1674, 68.45% [TWIST] in the nondisclosure group) after later learning about public health uses of ARVRT results. Conclusions: The majority of respondents expressed willingness to undergo ARVRT even with disclosure of public health uses of these data, but willingness markedly decreased when learning about these uses after the fact, highlighting the importance of transparency in MHS programs. Accordingly, in line with the ethical principle of respect for autonomy and the likelihood that the potential public health benefits of MHS programs will not be compromised, consideration should be given to encouraging clinicians to disclose public health uses of ARVRT at the time ARVRT is recommended.

Draft guidance from the U.S. Food and Drug Administration states that one can interpret a treatment effect on a clinical outcome assessment–based endpoint when expressed as some difference between group means. Recently, Trigg et al. examined different approaches for deriving thresholds for interpreting such between-group differences. In this commentary, I make several observations to advance further discussion around this issue. Some key points are (1) rather than “between-group difference,” specify the level at which you wish to infer a treatment effect: population or individual; (2) points of reference may be different for interpreting individual- and population-level treatment effect estimates; (3) who provides input and what types of anchor variables are used to generate points of reference might differ for interpreting individual- versus population-level estimates of treatment effect; and (4) in a parallel groups design, meaningful within-patient change is not especially relevant for understanding the meaningfulness of a treatment effect.