The CDC Morgellons Study: What It Found—and What It Couldn’t Prove

Executive summary

The short version

A note on the standard applied here. This commentary makes claims about study design—about who was enrolled, how they were selected, and what the sample size permits. Claims about design invite evaluation by design standards, and that is what follows. None of it is a comment on the author’s experience of illness, which is not in dispute.

A recent Lived Experience of Morgellons Disease commentary, “My Thoughts on the CDC Study”, reaches an important conclusion correctly: the 2012 CDC-sponsored investigation did not determine that Morgellons disease is delusional infestation. That correction is necessary. It is also where methodological precision must begin rather than end.

The commentary goes on to treat the study’s record-search code as a psychiatric enrollment criterion, regional scope as evidence of a biased sample, and the absence of a named Morgellons specialist as evidence of bias from inception. Those are different criticisms with different evidentiary requirements. They cannot be combined under the word bias and treated as self-proving.

A serious peer review identifies the design feature at issue, specifies which estimate it can affect, distinguishes imprecision from invalidity, and—where possible—states the likely direction and magnitude of the distortion. The STROBE statement for observational research makes the same basic demand: report participant selection, missingness, bias, precision, limitations and generalizability separately.

The Pearson study deserves neither the authority it has acquired nor wholesale dismissal. It should be read claim by claim.

Study design

What the study was built to measure

In its January 2008 announcement, CDC said the investigation would describe who might be affected, characterize symptoms and generate hypotheses about contributory factors. Pearson and colleagues later stated three objectives: characterize clinical and epidemiologic features, estimate prevalence in a defined population, and generate hypotheses about causative or contributory factors.

The design was a descriptive case series—a study that describes a group of patients without comparing them to anyone else—combined with a cross-sectional survey, clinical evaluations and histopathologic studies. It had no comparison group and no longitudinal follow-up capable of establishing temporal order.

That design could describe the people enrolled, estimate an operational prevalence within the defined Kaiser Permanente Northern California population, document findings in the specimens actually examined and generate hypotheses. It could not estimate risk-factor associations against controls, demonstrate causation, adjudicate a biological mechanism or settle whether every presentation called Morgellons shares one etiology.

Calling the study “not definitive” is therefore correct—but not a newly discovered fatal flaw. It is the scope the investigators declared, in a limitations section stating that the cross-sectional design and lack of a comparison group prevented assessment of risk factors and temporal relationships.

The paper also carries an explicit disclaimer that its findings do not necessarily represent the official position of CDC, KPNC or the Armed Forces Institute of Pathology. Whatever institutional authority later readers attached to the article, the publication itself did not announce an official CDC disease classification.

The scientific question is not whether the study was definitive. It plainly was not. The question is which observations remain valid inside its declared scope.

Back to navigation ⬆️

Denominators

Who was counted, and in what number

Most arguments about this study are really arguments about denominators. The study does not have one sample size. It has nested groups, and each group supports a different claim.

Read that table before reading any claim about what the study “showed.” Nearly every overstatement in either direction comes from attaching a finding to a larger group than the one it was measured in.

One distinction inside it matters. The passage from 70 survey participants to 41 clinical participants is attrition; the earlier reduction from electronically located records to 115 people meeting the case definition is principally screening and case adjudication, not the loss of confirmed cases. A sensitive keyword search is expected to generate false positives, and removing people who do not meet the prespecified definition is part of ascertainment.

Attrition can introduce bias when participation is related to the findings of interest. Pearson et al. reported that survey completers were more likely to be female than noncompleters but did not differ by age, and that participants receiving clinical evaluation were demographically similar to survey completers who did not, although the underlying data were not shown. Those comparisons reduce—but do not eliminate—concern about nonresponse bias. Similarity on age and sex does not establish similarity in lesion type, illness severity, psychiatric measures or material composition.

Back to navigation ⬆️

Physical findings

“No common source” does not mean “nothing physical was found”

The commentary paraphrases the abstract as finding no “infectious or other physical source.” That wording is broader than the paper’s conclusion. Pearson et al. reported that they identified no common underlying medical condition or infectious source capable of explaining the group.

The study did find physical abnormalities. Among the 41 clinically evaluated participants, 31 were judged to have either biopsy-suitable lesions or material available for collection. Across the clinical phase, investigators obtained 62 biopsies: 37 lesional, 22 from clinically normal skin and three from undocumented sites. Among the 37 biopsied lesions, 19 (51%) showed solar elastosis, 15 (40%) showed features of excoriation or chronic irritation, and six (16%) had findings consistent with an arthropod bite or drug allergy. Special stains detected gram-positive bacteria in 12 specimens from 11 participants and fungi in eight specimens from eight participants. Culture swabs were obtained from 28 case-patients with open or purulent lesions; organisms grew from the lesions of 15, which the investigators interpreted as secondary infection. Those findings did not converge on a single primary cause.

This distinction is not semantic housekeeping:

Failure to identify one cause shared by a cohort is not equivalent to finding no physical pathology in its members.

The same rule applies in the opposite direction. Finding inflammation, secondary infection or an ordinary material in one participant does not establish the cause of the full syndrome—or the cause in another participant.

Back to navigation ⬆️

Case ascertainment

The search code was not the case definition

The most consequential error in the new commentary appears in its account of case ascertainment. It states:

“Cases were identified through electronic health records searching for these types of words in the progress notes or were coded in the record as ICD-9-C code 300.29, which translates to delusions, parasitosis.”

and concludes that the investigators “only included those who were diagnosed as such in their prevalence analysis.”

That is not the procedure reported in the paper.

Pearson et al. used three distinct stages:

1. Electronic search: records were located if selected words—such as Morgellons, fiber, thread, fuzzball, dots, specks, granules or delusion—appeared in progress notes, or if ICD-9-CM code 300.29 appeared. The authors state that 300.29 was the code used at KPNC for patients with Morgellons.
2. Medical-record review: investigators applied predefined criteria to determine whether the located records described suggestive signs and symptoms.
3. Standardized telephone screening: potential participants were tested against the study’s operational case definition before enrollment.

The case definition required a report of fibers or other solid material emerging from the skin plus a skin lesion, a disturbing skin sensation, or both. It did not require a diagnosis of delusional parasitosis. It did not require a belief in parasites, a fixed false belief or persistence of that belief despite contrary evidence.

The word or in the electronic search matters. A record containing fiber could be found without code 300.29; a record carrying the code could be found without another keyword. The code and keywords generated candidates. Neither replaced the later case-definition screen.

There is a further point the commentary misses, and it cuts in the commentary’s own favor. The claim that 300.29 “translates to delusions, parasitosis” repeats Pearson’s parenthetical gloss as though it were the code’s definition. It was not. In the contemporaneous ICD-9-CM official code titles, 300.29 is “other isolated or specific phobias.” Delusional disorder was coded 297.1.

ICD-9-CM 300.29 was not formally a delusional-parasitosis code. Pearson reports that KPNC used it locally for Morgellons. Its presence could locate a record. It cannot establish that the patient was diagnosed with delusional parasitosis.

That is a stronger and more defensible statement than the one the commentary makes. It does not dissolve the underlying concern. A nonspecific residual code applied by local institutional convention could still have shaped who entered the case pool—patients whose clinicians documented none of the searched words could be missed—and the paper does not report how many of the 104 record-identified cases were found through code 300.29, through nonpsychiatric keywords or through both. The effect of the code on case capture and cohort composition therefore cannot be quantified from the published data.

That is a defensible limitation. It is not evidence that a delusional-parasitosis diagnosis was required.

Back to navigation ⬆️

Prevalence

Why the 11 self-identified cases were excluded from prevalence

Eleven KPNC members who contacted the study independently also met the same symptom-based definition and were allowed to participate. They were excluded only from the prevalence numerator.

That decision is methodologically coherent. The denominator was the KPNC population under systematic electronic surveillance, and self-referral introduced a different, unmeasured probability of capture. Adding self-selected cases to a numerator generated by one ascertainment system, while retaining a denominator defined by another, would tend to inflate the estimate.

Accordingly, the published prevalence—3.65 cases per 100,000 enrollees, 95% confidence interval 2.98–4.40—used the 104 electronically identified cases against an average monthly KPNC enrollment of 2,850,606. Two population figures appear in the paper and should not be read as a discrepancy: KPNC covered approximately 3.2 million enrollees across 13 Northern California counties, while 2,850,606 was the average monthly enrollment used as the prevalence denominator for the study window.

The estimate did not purport to measure worldwide Morgellons prevalence; it estimated the frequency of cases detected by this operational definition and ascertainment procedure in that source population. It is vulnerable to missed documentation, care-seeking patterns, the English-language criterion, uneven implementation of electronic records and possible phenotype misclassification. Those limitations bound its interpretation. They do not make the arithmetic nonexistent.

Back to navigation ⬆️

Generalizability

Geographic restriction affects transportability, not automatically internal validity

The commentary argues:

“All of these individuals are from a very small geographic area and the condition being studied is global. This leads to the likely possibility of a biased sample.”

Geographic scope and selection bias are not synonyms. Transportability is the question of whether a result travels to other populations. Internal validity is the question of whether the result is correct for the population actually studied. They fail independently.

KPNC then covered approximately 3.2 million people—nearly 30% of the population across 13 counties. That is a large defined sampling frame. The study’s regional nature plainly limits transportability: its demographic pattern and prevalence estimate cannot simply be applied to South Carolina, Europe or the world.

But a regional source population is not biased merely because it is regional. To establish selection bias, one must show that inclusion in KPNC, presentation for care, documentation or study participation was associated with the relevant phenotype in a way that distorted an estimate. That is plausible and worth investigating. It is not demonstrated by pointing to the borders of the catchment area. The precise conclusion is narrower: the study provides a KPNC-specific estimate under a particular case-finding system, not a global prevalence estimate.

Back to navigation ⬆️

Statistical precision

“Small sample” is not a universal nullifier

The commentary writes:

“This small sample size for a condition that is so poorly understood does not allow for any clear statements about cause or even symptoms.”

That sentence combines two very different inferential problems. Sample size primarily controls precision—how tightly a number is pinned down. Selection and measurement processes control validity—whether the number measures the right thing at all. The absence of a control group prevents causal inference. It does not erase observed symptom frequencies.

In plain terms: with 70 people, a finding reported by about half of them is pinned down to roughly a 39%–61% range. With 12 people, the same finding is compatible with anything from about a quarter to three-quarters of the wider group. The 12-person material subset is therefore poorly suited to precise population-wide composition claims, while the 70-person survey can still describe what those 70 participants reported, with stated uncertainty.

Back to navigation ⬆️

Construct validity

The broad phenotype is a real limitation—but its direction cannot be assumed

The CDC case definition was substantially broader than a definition requiring objectively documented filaments embedded in or projecting from intact tissue. Of the 115 confirmed case-patients, 81 described fibers, alone or with other material; 34 described other forms of material, including specks, granules, dots, worms, sand, eggs, fuzzballs or larvae.

If a biologically distinct disorder is defined specifically by tissue-integrated filaments, combining that phenotype with surface debris, nonfiber material and heterogeneous lesions could dilute a real signal. This is a legitimate concern about construct validity and case misclassification.

It does not follow that all participants were psychiatrically selected, nor that a clinician’s impression of a “true Morgellons” presentation should replace a reproducible definition. Selecting participants because a recruiting clinician already judges their fibers authentic can create incorporation bias: the disputed finding becomes both the entry criterion and part of the conclusion.

The solution is prospective stratification, not allegiance. A stronger study would separately define and analyze filaments objectively documented in or projecting from unbroken tissue under a prespecified imaging protocol, and material present only on the skin surface, in crust or in self-collected specimens—then compare both groups with matched dermatologic, delusional-infestation and healthy controls. That design could test whether the supposedly narrower phenotype predicts distinct histology, chemistry, microbiology or clinical features.

Back to navigation ⬆️

Clinical methods

Clinical familiarity is useful; it is not a substitute for blinding

The commentary notes:

“There is no mention of a physician within the group with clinical experience specifically examining Morgellons filaments or the disease itself.”

That may matter if experienced clinicians recognize a reproducible phenotype that general dermatologists miss. The claim should be tested, not dismissed.

But the study did not lack relevant clinical or laboratory expertise. An internist performed medical examinations; a dermatologist performed separate skin examinations and collected material; a second dermatologist independently reviewed reports, photographs and pathology; infectious-disease pathologists evaluated biopsy sections; and two dermatopathologists blinded to the clinical diagnosis evaluated H&E sections at the Armed Forces Institute of Pathology. Absence of a named Morgellons-treating clinician is therefore not equivalent to absence of dermatologic, pathologic or materials expertise. Conversely, adding one advocate-aligned clinician would not repair the lack of controls or establish an etiology.

One qualification belongs here rather than in the study’s favor. Blinding examiners to an individual patient’s clinical diagnosis is not the same as blinding them to the study’s framing. Reviewers who know they are evaluating specimens from an investigation of an unexplained dermopathy discussed in the literature as delusional infestation are not hypothesis-blind, whatever the case labels say.

The stronger design is to use disease-specific experience to construct an explicit protocol, then test that protocol with masked examiners and independent laboratories. Expertise should generate reproducible criteria. It should not function as an unblinded gold standard.

Back to navigation ⬆️

Mechanistic inference

The study examined biology, but it did not test a mechanism

The commentary says the investigation included no study of potential biological mechanisms. That is partly right and partly too broad.

Pearson et al. did not conduct a mechanistic experiment capable of showing how a causal agent produces lesions, sensations or filaments. No comparison group, temporal sequence or treatment experiment connected an exposure to a biological pathway and then to the phenotype. The investigation nevertheless included extensive biological observation: clinical examination, lesional and normal-skin biopsy, H&E and special stains, bacterial and fungal culture, conditional immunohistochemistry and PCR, serology, polarized-light microscopy, infrared spectroscopy and SEM/EDXA. Calling that “no study” of biology understates what was done; calling it a demonstration of mechanism would overstate it. The accurate description is broad phenotyping and exploratory etiologic testing, without a design capable of establishing a mechanism.

Back to navigation ⬆️

Lyme testing

The Lyme criticism identifies the wrong missing variable

The commentary argues that “the study does not state whether the patients were asked explicitly about exposure to B. burgdorferi,” and that “it seems remiss that patients would not be asked.” Exposure history could have been useful—particularly tick exposure, prior erythema migrans, earlier Lyme diagnoses, antibiotic treatment and the timing of those events.

But a participant cannot reliably report microbial exposure itself. A recalled tick bite or previous diagnosis is not evidence that B. burgdorferi was present, while absence of a recalled bite does not exclude exposure.

The investigators did test serum from clinically evaluated participants using an EIA and IgG Western blot. One case-patient had a positive EIA and one a borderline EIA; none had a positive IgG Western blot. The authors used IgG because, in the paper’s wording, serum specimens were collected “130 days after illness onset.” The paper states that figure flatly, without specifying whether it is a minimum, median or mean; it is quoted as published rather than given a precision the reporting does not have. Current CDC guidance still recommends two-step serology while acknowledging important limits: tests can be falsely negative early, early antibiotic treatment can reduce seroconversion, and antibodies can persist for years.

Pearson’s result therefore supports a narrow statement: the clinically evaluated subset did not show CDC-defined IgG seropositivity under the assay and threshold used. It does not establish that no participant had ever encountered Borrelia, or that Lyme disease is absent from every population labeled Morgellons.

A more consequential limitation is that the study was not a prespecified, adequately powered test of a Borrelia-Morgellons hypothesis. The biopsy pathway used special stains in specimens with inflammatory infiltrates and proceeded to IHC or PCR when an infectious agent was identified; it did not report systematic Borrelia-specific molecular testing of every relevant lesion with matched controls. That is the gap a pathogen-specific replication would need to address.

Back to navigation ⬆️

Comparative evidence

Savely and Stricker describe a different cohort, not a corrective control group

The commentary cites Savely and Stricker’s 2010 report, a descriptive series of 122 patients selected for clinically observed microscopic subcutaneous fibers. That paper matters because it operationalized a narrower filament-centered phenotype than Pearson’s later study.

It does not establish that Pearson’s participants were the wrong patients.

Savely and Stricker examined patients from one clinician’s practice who already met the defining observation. The study had no matched control group, did not estimate prevalence from a defined population and was not designed to show that the observed filaments were nontextile by systematic chemical comparison. It can describe the selected cohort and motivate a narrower case definition. It cannot determine how common that phenotype is among everyone reporting Morgellons, how often similar microscopic findings occur in controls, or whether one etiologic process produced the fibers.

The two studies sampled different constructs: Pearson broadly sampled a symptom-defined unexplained dermopathy within a healthcare population, while Savely and Stricker narrowly described a clinician-selected, filament-positive practice cohort. Neither adjudicates the other. A direct comparison under a common prospective protocol would.

Back to navigation ⬆️

Bias analysis

“Bias” must name a mechanism

The commentary concludes:

“They were biased towards a psychiatric cause. This fact, in and of itself, mandates that additional research be done.”

There are reasons to scrutinize that possibility. The introduction foregrounded the prevailing delusional-parasitosis interpretation; delusion and code 300.29 appeared in electronic case finding; neuropsychiatric tests received substantial attention; and the abstract compared the findings with delusional infestation.

None of those observations, alone, demonstrates that the results were generated by psychiatric selection. Calling bias a fact is where the argument fails: bias is a mechanism that has to be specified and, where possible, measured. The operational case definition did not require a delusion; dermatopathologists were blinded to clinical diagnosis; the laboratory program investigated medical, infectious, environmental, toxicologic, neurocognitive and psychiatric possibilities; and the 2008 public protocol announced mental-health assessment as one component of a broader examination, not as the predetermined answer.

A rigorous bias claim would ask:

1. Selection bias: Did psychiatric coding make people with that diagnosis more likely to enter the case pool than filament-positive people without it? The paper does not provide the search-route counts needed to quantify this.
2. Information bias: Were lesions, fibers, exposures or beliefs measured differently because examiners expected a psychiatric result? Standardized forms, independent review and partial blinding reduce some risks but do not eliminate them—and blinding to diagnosis is not blinding to hypothesis.
3. Misclassification: Did the broad case definition combine biologically different presentations? Quite possibly; the 34 nonfiber reporters make this concern concrete.
4. Confounding: Did drug use, dermatologic disease, psychiatric morbidity, treatment history or care-seeking behavior account for observed associations? Without controls, the study could not resolve this.
5. Selective interpretation: Did the abstract or discussion extend beyond the results? The phrase “similar to” is cautious. What followed was not.

This framework does more than attach the word bias. It shows what evidence would confirm or reduce each concern.

Back to navigation ⬆️

Evidence boundary

What the study actually contributes

Read at the correct resolution, Pearson et al. supports several limited conclusions:

• A symptom-defined unexplained dermopathy was uncommon under the study’s ascertainment system in the KPNC population.
• Participants reported substantial impairment and heterogeneous cutaneous and multisystem symptoms.
• The evaluated lesions were heterogeneous; many showed chronic irritation or excoriation, and some had secondary infection.
• The loose-material analysis was small: 23 specimens from 12 participants. Of those specimens, 83% contained protein—interpreted as likely superficial skin—and 43% contained cellulose consistent with cotton fibers, some with evidence of dyes. These categories overlapped, and the percentages cannot be generalized to all 115 case-patients.
• Three specimens contained additional materials: polyamide (probably nylon), cellulose nitrate containing bismuth (consistent with nail polish), and polyethylene, which the investigators identified as a possible contaminant from the specimen container lid.
• In the clinically evaluated subset, 23 of 39 participants (59%) showed impairment in at least one cognitive domain, 25 of 40 (63%) had a clinically significant elevation on a Personality Assessment Inventory clinical domain—most often the somatic-complaints scale—and 20 of 40 (50%) had at least one drug detected in hair. These are observations requiring controls and clinical context, not diagnoses. The authors themselves said they could not determine whether detected drug use contributed to symptoms, was used in an attempt to relieve them or was unrelated.
• The evaluated group did not reveal one medical condition or infectious source shared across participants, and the design did not determine whether the presentation was a new condition or delusional infestation.

That container finding deserves emphasis, because it is the study’s own evidence for a methodological point made below. In an investigation whose central object is material recovered from skin, one of the 23 analyzed specimens contained a polymer the investigators themselves flagged as possibly originating in their own collection container. Without container blanks, field blanks and textile controls, a future study could not reliably distinguish patient-associated material from contamination introduced during collection or processing.

The study does not support saying that CDC proved Morgellons is delusional parasitosis, that all fibers were cotton, that 115 people underwent fiber analysis, that no physical abnormalities existed, or that all presentations called Morgellons have one cause. It also cannot be reduced to “nothing useful” because later readers overstated it.

Back to navigation ⬆️

Downstream interpretation

How the study was later overstated

That overstatement is not hypothetical, and it is not confined to comment sections. DermNet—a dermatology reference consulted by clinicians and patients—tells readers, in a version last reviewed in December 2024, that the CDC study concluded Morgellons to be “a subtype of delusional Infestation.”

Pearson et al. did not reach that conclusion. They wrote that they could not determine whether the condition was new or represented wider recognition of an existing condition. The same passage reports 113 cases—neither the 115 who met the case definition nor the 104 identified electronically—and dates the study to 2008, the announcement year, not the 2012 publication.

This is how a hedged sentence in a limitations-heavy paper becomes a settled classification in the reference literature. That does not place Pearson’s paper beyond criticism. Its framing, its incomplete ascertainment reporting and its declared limitations remain fair targets, and this article has pressed several of them. What it means is narrower—the specific claim that CDC classified Morgellons as delusional infestation is a downstream reinterpretation rather than the investigators’ stated conclusion.

Back to navigation ⬆️

Research blueprint

What a decisive next study would require

A stronger investigation would separate prevalence, phenotype and mechanism rather than asking one case series to answer all three.

1. Prospective, multicenter recruitment. Recruit from dermatology, infectious disease, psychiatry, primary care and community referral across several regions, recording the source and applying one screening protocol to everyone.

2. Prespecified phenotype strata. Distinguish reported surface material from filaments documented in intact tissue, recording lesion morphology, filament location, tissue continuity and specimen provenance before laboratory results are known.

3. Appropriate controls. Include matched participants with inflammatory or pruritic dermatoses, excoriation disorders, diagnosed delusional infestation and healthy skin. A material is not disease-specific merely because it is found in cases.

4. Controlled collection and chain of custody. Photograph and map sites before manipulation, then use standardized cleansing procedures, clean instruments, field blanks, container blanks, textile controls and documented custody from bedside to laboratory. Pearson’s own polyethylene finding shows why this is not a formality.

5. Orthogonal tissue and material analysis. Combine histology with connective-tissue stains and immunostains, polarized microscopy, FTIR or Raman spectroscopy, SEM/EDX and proteomics where feasible. Predefine how each material class will be classified, and report the unidentified fraction rather than forcing every specimen into a preferred category.

6. Pathogen testing that can survive contamination review. Any sequencing or molecular testing needs extraction blanks, reagent controls, positive controls and blinded sample identifiers, with candidate organisms confirmed by an independent method and laboratory. Detection must be linked anatomically and statistically to the phenotype before causation is proposed.

7. Blinded assessment across domains. Dermatopathologists, materials scientists, microbiology laboratories and psychiatric evaluators should be masked to one another’s results. A psychiatric diagnosis should not be inferred from a specimen report, and a pathogen result should not determine whether a filament is judged tissue-derived.

8. A statistical plan matched to the question. Prespecify primary outcomes, power the study for between-group comparisons, account for multiple testing, and publish the participant flow by recruitment route and phenotype.

This design could reveal one disorder, several disorders sharing a label or overlapping processes. Heterogeneity is an empirical possibility, not an excuse to avoid classification.

Back to navigation ⬆️