What Screening Evidence Actually Has to Show
The 2026 AHRQ review of medical care for adults with Down syndrome offers a useful test of evidence discipline. Its central question is not simply which conditions are common, or whether a test can find an abnormality, but whether screening and diagnosis lead to better outcomes than another strategy or no screening. Understanding that distinction helps the reader classify what a study actually supports, interpret the review’s empty screening evidence base, and avoid treating uncertainty as proof that care is ineffective.
Four questions hidden inside the word “screening”
A screening discussion often starts with a true observation and ends with a claim that the observation cannot carry by itself. Suppose a condition is more prevalent among adults with Down syndrome. That finding can justify attention, research, or a conversation about prevention. It does not yet show that routinely testing people without symptoms will improve their health. To reach that conclusion, several separate questions have to be answered.
The first is epidemiologic: how often does the condition occur, and in whom? A prevalence estimate can reveal age patterns, differences from the general population, or a group that may deserve closer attention. It can also be distorted by how people enter the study. Electronic health records tend to capture recognized and coded diagnoses. Specialty clinics may see people with more complex problems. Direct examinations may find conditions that routine records miss, but a referred sample may not represent all adults with Down syndrome. The number is therefore useful only with its denominator, setting, age range, diagnostic method, and comparison group.
The second is diagnostic: can a test or assessment identify the condition accurately in the population and circumstances where it will be used? A tool may perform well in a research protocol but be difficult to complete in an ordinary primary-care visit. Communication differences, the need for cooperation, sedation, or a specialist’s interpretation can change its practical performance. Even an apparently accurate result can have a different meaning when the underlying condition is common, uncommon, or expressed differently from the population in which the tool was developed.
The third is therapeutic: once a condition is identified, is there an intervention that improves an outcome that matters to the adult? “Treatment is available” is not a sufficient answer. The treatment must be tolerable, accessible, and effective for adults with Down syndrome, not merely for a different population. The outcome also matters. A change in a laboratory value may be useful, but it is not the same as fewer complications, better function, improved quality of life, or reduced caregiver burden.
The fourth is programmatic: what happens when the test is offered routinely to a defined population? This question includes the whole sequence from invitation to follow-up. A screening program may identify disease earlier, but it may also produce false positives, additional tests, anxiety, cost, medical trauma, unnecessary medication, or restrictions. Some people who would never have developed a consequential illness may be labeled and treated. Others may receive a reassuring result and delay seeking care when symptoms later appear. Program effectiveness is consequently a comparative outcome, not a count of abnormalities detected.
The AHRQ review organized these distinctions into two key questions and two contextual questions. Its first key question, KQ1, asked about the benefits, harms, and considerations of screening and diagnostic interventions for co-occurring medical and behavioral-health conditions in adults with Down syndrome. The second, KQ2, asked about interventions used to treat those conditions. A contextual question addressed which conditions were more or less prevalent and how estimates varied by characteristics such as age or sex. Another asked how common behavioral and mental-health conditions presented in adults with Down syndrome compared with the general adult population.
This architecture prevents evidence from one question from being quietly substituted for evidence from another. A prevalence study can inform the first step of the reasoning, but it cannot answer whether screening improves outcomes. A treatment trial can show what happened after people were already identified, but it does not show whether testing asymptomatic people is worthwhile. A validity study can describe how a tool performs under particular conditions, but it does not measure the balance of benefit and harm produced by a screening policy.
The review’s logic model makes the expected pathway visible. Screening or diagnosis should lead to earlier detection, accurate diagnosis, useful prognostic information, or access to beneficial treatment. Treatment should then affect symptoms, function, quality of life, mortality, or caregiver well-being, while harms are tracked alongside benefits. A missing link does not make the other links irrelevant. It does mean that the final conclusion must stop before claiming a benefit that has not been measured.
The difference between these questions is especially important in adults with Down syndrome because the evidence may be strong in one part of the chain and weak in another. The population can have a higher prevalence of a condition, while the appropriate threshold for testing remains uncertain. A test can identify change, while the cause of that change remains unclear. A diagnosis can be clinically important, while the benefit of finding it before symptoms emerge has not been evaluated. A recommendation can still be reasonable, but its justification should be described as a balance of evidence, harms, feasibility, and values rather than as a completed proof of net benefit.
What the empty screening literature does—and does not—tell us
The AHRQ review searched MEDLINE, Embase, PsycInfo, and CINAHL for English-language studies published from 2000 through July 10, 2025. It also examined reference lists, ClinicalTrials.gov, public suggestions, and selected gray literature. For its key questions, it required studies of adults with Down syndrome using randomized or appropriate comparative designs. For screening and diagnosis, the comparator could be another test or no screening. Studies about implementation alone were excluded, as were case reports and case series that could not support a comparative effect estimate.
After reviewing 8,680 unique references, the investigators identified 42 eligible publications representing 36 unique studies. None addressed KQ1. The 36 studies all concerned treatment, and 52 publications supplied selected prevalence evidence for the contextual question. The result is striking because the search was not asking whether clinicians ever screen adults with Down syndrome, whether a test has been described, or whether a condition is biologically plausible. It was asking for comparative evidence that screening or diagnosis changes outcomes in this population. No study met that standard.
That finding should be stated precisely. It means that the review could not estimate the benefits and harms of comparative screening or diagnostic strategies for the conditions it examined. It does not mean that every screening test is inaccurate. It does not mean that a symptomatic adult should wait for a screening trial before receiving an appropriate diagnostic evaluation. It does not mean that the absence of an eligible study proves that screening has no benefit. It identifies a gap between clinical concern and evidence of program-level effect.
The distinction also explains why the review did not treat the 36 treatment studies as indirect proof of screening benefit. Those studies addressed what happened after an intervention was offered, often to a selected group. Sixteen concerned dementia, five concerned obesity, and three each addressed behavioral conditions, cardiovascular disease, bone, joint, and mobility conditions, sleep apnea, or vision. No treatment studies were identified for blood cancers, celiac disease, diabetes, solid tumors, or thyroid conditions in adults with Down syndrome. Treatment evidence cannot fill a missing screening comparison merely because treatment is downstream from diagnosis.
The included treatment literature was also difficult to interpret as a whole. More than half of the studies were assessed as having high, critical, or serious risk of bias. Many were small: ten had 25 or fewer participants, and only three recruited more than 200. Participants, interventions, outcome measures, and follow-up periods varied enough that the review could not combine the results in a quantitative or conceptual meta-analysis. Some studies reported favorable signals—for example, possible improvements in aspects of dementia, depression, sleep apnea, or mobility—but the report rated the evidence insufficient for definitive conclusions.
“Insufficient” has a technical meaning here. AHRQ considered study limitations, consistency, directness, precision, and reporting bias. Evidence is direct when the link between intervention and outcome is close to the question of interest. It is precise when the estimate is narrow enough to exclude clinically different interpretations. It is consistent when studies point in a similar direction and magnitude. If a body of evidence is small, unreplicated, biased, indirect, or too imprecise, a favorable result may be a hypothesis rather than a reliable effect estimate.
This is why statistical significance alone cannot settle the question. A small study can produce a low p-value by chance, through selective measurement, or because the measured outcome is only a proxy for what matters. A larger study can produce an imprecise result if events are rare or follow-up is incomplete. The reader needs to know who was compared, what outcome was measured, how long participants were followed, and which assumptions connect that outcome to health. The AHRQ report’s strength-of-evidence framework is an attempt to keep those assumptions visible.
Prevalence and test-validity findings nevertheless remain important. They can identify conditions that deserve attention, reveal age patterns, expose a possible mismatch with general-population guidance, and suggest which tools or outcomes need better study. The review describes prevalence evidence drawn mainly from retrospective records using diagnostic codes, but also notes direct methods such as cognitive assessments and neuroimaging for dementia, polysomnography for sleep apnea, dual-energy X-ray absorptiometry for osteoporosis, biopsy for celiac disease, and ocular examinations for vision. These methods do not measure the same thing. A record-based estimate may be lower than an examination-based estimate because undiagnosed disease is invisible to the record; a sleep-study estimate from referred participants may be higher than a population estimate because referral is not random.
The review also found validity research on diagnostic tools, especially for dementia. Those studies can help researchers ask whether a caregiver questionnaire, cognitive assessment, or biomarker identifies a meaningful stage of disease. But because they did not meet the comparative KQ1 criteria, they do not show that using the tool routinely improves clinical outcomes. A tool can be valid for detecting a signal and still have uncertain value as a screening program if the next diagnostic step is burdensome, the result is difficult to interpret, or no treatment has proven benefit at that stage.
The most responsible conclusion is therefore neither “screen everything” nor “screening is unsupported and should be abandoned.” It is that the evidence currently supports different levels of confidence for different claims. Epidemiology can justify vigilance. Diagnostic research can refine how a signal is investigated. Treatment studies can identify promising options. Comparative screening studies are needed to show whether offering a test routinely improves outcomes after benefits, harms, access, and follow-up are included.
That separation makes the next clinical comparisons possible. Dementia, thyroid disease, diabetes, and sleep apnea all raise legitimate concerns in adults with Down syndrome, but they do not share the same symptoms, tests, treatment pathways, or evidence gaps. High prevalence may lead to active surveillance in one case, periodic laboratory testing in another, and a more conditional approach in a third. The difference is not inconsistency. It is what evidence-based reasoning looks like when the links between risk and outcome are not equally strong.