Journal of Evidence and Innovation | Volume 2, Issue 10 | Published 2026-10-29
Broad graft labels combine properties that may have different biological and clinical implications. Source, processing, storage, geometry, and delivery can vary together, making an observed outcome difficult to attribute to one material feature. This conceptual synthesis develops an exposure definition for superior capsular reconstruction. A review of graft biology motivates the distinction, while reconstruction research and causal methods clarify the evidence needed for comparison. The framework separates source identity from measured construct properties, places processing on an explicit timeline, and examines overlap among the strategies available to patients. It treats biological plausibility as a hypothesis-generating resource rather than a substitute for comparative outcomes. No specimens, operative records, or patient outcomes are newly analyzed. The contribution is a design for describing what was implanted and what a clinical comparison can legitimately estimate when nominal graft categories contain multiple treatment versions.
Keywords: superior capsular reconstruction; graft source; material processing; treatment versions; biological incorporation.
A material category can be useful for communication while remaining too broad for causal interpretation. Two constructs described with the same source label can differ in processing, dimensions, storage, or the conditions under which they were implanted. A later outcome difference may therefore concern more than the source itself. The first research task is to identify the object delivered to the patient at a resolution that matches the proposed explanation. Without that description, a precise comparison can attach certainty to a label whose underlying intervention is unclear.
Cao and colleagues' review makes graft biological properties a central question in superior capsular reconstruction (Cao et al., 2025). Its value for the present framework is to motivate separating a material name from the features through which incorporation might occur. The review does not supply a new randomized source comparison for this article. We use its question to design a study record that could test a specified relationship. The proposed record preserves source and processing as related exposures while leaving their comparative clinical consequences open to observation.
Clinical research on superior capsular reconstruction places graft use inside a complete operative pathway (Mihata et al., 2013). That pathway includes patient selection and a constructed mechanical environment, not merely an isolated sheet of material. A source-based interpretation should therefore retain the context in which the graft was used. The same principle applies to a null result: absence of a group difference under one mixture of processing and delivery does not establish biological equivalence for every construct bearing the same source label. The treatment contrast needs to remain visible in either direction.
Source records identify origin, but origin is not a direct measurement of every biologically relevant property. Cell content, extracellular structure, hydration, and mechanical behavior require their own observations when they form part of the hypothesis. Tissue-engineering research provides a foundation for treating a scaffold and its interaction with a host as a system (Langer and Vacanti, 1993). It does not make a source category a sufficient description of that system. A study should state which properties were measured and which were inferred from the material's name or documentation.
A useful data structure can retain three layers: the supplier or collection description, the processing history, and the measurements obtained on the construct or an appropriately linked sample. Those layers should not be merged silently. A measurement from one specimen may describe a batch incompletely, and a batch specification may not establish the state of the particular implanted object. The record should preserve the sampling relationship. This allows an analyst to distinguish an observed property from a default assumption assigned to every patient receiving that category.
Geometry also belongs in the exposure description. Thickness, dimensions, orientation, and the area available for interface contact can affect both a mechanical test and the clinical construct. A comparison of source categories that systematically differ in geometry may estimate a package of changes. That can be a meaningful strategy contrast if stated explicitly. It should not be described as the isolated effect of origin unless the design supports that narrower attribution. A study can preserve both the broad label and its measured versions without pretending that one replaces the other.
Processing can occur before storage, during preparation, or immediately before implantation. Its timing determines whether a later measurement represents the delivered construct. A research record should identify which steps are documented, which are site-specific, and which remain unknown. An example such as an undocumented change in preparation illustrates the problem without asserting that it occurred in a particular study. If two sites use the same product name but follow different preparation pathways, the source label alone cannot establish comparable exposure.
Measurements can themselves alter a specimen, so direct characterization of the implanted object may not be feasible for every property. A companion sample can support a measurement when its relationship to the implant is defined. Investigators should report how that sample was selected and whether storage or preparation differed. The resulting uncertainty should remain attached to the inferred property. A detailed assay does not become an exact implant measurement merely because its numerical output has many decimal places. Precision of the assay and representativeness of the sample are separate questions.
Processing and source may interact. A step intended to preserve one property can affect another, and a procedure can have different consequences for different starting materials. A prospective design should identify the interaction it wishes to test rather than examine every possible combination after seeing the outcomes. Sparse combinations can make estimates unstable. Reporting the treatment versions and their frequencies is therefore an important first result. It can reveal that the desired source comparison is unsupported before investigators invest in an elaborate adjustment model.
A causal diagram can locate source, processing, patient selection, and outcomes on a common timeline (Greenland et al., 1999). A processing choice may be part of the intervention, a response to a baseline constraint, or a variable tightly linked to site. Its role cannot be decided solely from the column name. The investigator should state whether the question concerns replacing one source under standardized preparation or choosing among complete strategies as ordinarily delivered. These estimands require different interpretations and can legitimately produce different answers.
Clinical eligibility should identify patients for whom the compared strategies were plausible. A source may be unavailable, unsuitable under documented constraints, or simply not preferred by a team. Combining those states conceals why treatment overlap is absent. The record should preserve the alternatives considered where feasible. A comparison restricted to patients with plausible alternatives can apply to a narrower population but have a clearer meaning. A larger comparison that relies on unsupported extrapolation is not automatically a stronger study because it includes more records.
Propensity scores organize measured treatment-assignment information under specified assumptions (ROSENBAUM and RUBIN, 1983). Their role is not to prove that all relevant selection factors were observed. A strong assignment classifier can indicate that groups are difficult to compare rather than that adjustment will be successful. Balance and overlap should be examined in relation to the intended contrast. A source category used almost exclusively in one clinical context may leave little support for the counterfactual comparison, regardless of how accurately a model predicts its use.
Matching methods offer another way to construct a comparison among observed patients (Stuart, 2010). The report should show who was matched, who was excluded, and which population the resulting estimate represents. Matching can change the question from a broad source comparison to a contrast among a specific subset with shared characteristics. That change can improve interpretability when disclosed. It should not be hidden behind the original cohort's sample size. The remaining unmeasured clinical judgments still require discussion even when measured balance looks satisfactory.
A biological hypothesis should predict a defined outcome at a stated stage. Interface observations, structural integrity, symptoms, and further procedures are related without being interchangeable. The Shoulder Pain and Disability Index provides a defined patient-reported domain (Roach et al., 1991). It can be part of a clinical comparison without directly measuring cellular incorporation. The study should identify why a proposed source or processing difference is expected to influence that domain and what intermediate observations would strengthen the explanation. A favorable questionnaire change cannot fill every missing biological link.
Repeated assessments should preserve the observation window and the reason for collection. A symptom-triggered image and a scheduled examination represent different ascertainment processes. If one source group is assessed more intensively, an apparent difference can partly reflect detection. Similarly, a participant who receives additional treatment may stop contributing measurements of the original construct. The record should retain these transitions rather than replacing them with one final success label. The outcome should describe the clinical course actually observed, including what remains unknown.
Heteroskedasticity-consistent covariance estimation addresses a variance assumption in a specified regression (White, 1980). It does not validate the exposure label, remove selection bias, or make missing outcomes ignorable. A narrow robust interval can therefore surround a poorly identified source contrast. Statistical reporting should remain connected to the design. The analyst needs to explain which uncertainty concerns sampling under the model and which concerns treatment classification or the observation process. Additional decimal places do not compensate for an unmeasured difference in what was implanted.
ROBINS-I provides a framework for examining bias in nonrandomized intervention studies (Sterne et al., 2016). For this question, treatment classification deserves attention alongside confounding. A broad source code can merge processing versions in a way related to site, calendar period, or patient selection. The assessment should connect that problem to the intended interpretation. It is not resolved by assigning one overall quality score. A useful review can show where exposure detail is sufficient for a source comparison and where only a broader strategy description is justified.
STROBE supports transparent reporting of observational populations, variables, and analyses (von Elm et al., 2007). The exposure record should add construct identity, documented processing, the sampling basis of measurements, and changes over time. The report should distinguish unavailable information from a measured absence. An unknown processing feature should not be replaced by the common specification without disclosure. Retaining that uncertainty can make a comparison narrower, but it makes the evidentiary chain more faithful to the source documentation and easier for another investigator to assess.
A reproducible study also needs versioned definitions. A product description or local preparation protocol can change while the broad source code remains constant. Combining those versions without recording the change can create an artificial temporal pattern. A research dataset should therefore preserve a link to the definition used for each episode. Harmonized analysis variables can be derived later, with the mapping retained. This keeps a useful simplified category from erasing the differences needed to interpret the original observations or reproduce an earlier analysis.
A prospective comparison could begin with a compact construct form and a small set of predefined material measurements on linked samples. It would record the reason for source selection, the processing steps that are actually documented, and the clinical outcomes relevant to the hypothesis. Feasibility should be tested before expanding the form. An exhaustive list that teams cannot complete reliably may be less useful than a focused core with clear definitions. The design should preserve why detailed measurements are available for some episodes and not others.
The analysis plan would state the population and the treatment versions before selecting an adjustment method. It would show the combinations that are observed and identify which comparisons rely on extrapolation. Mechanistic substudies could then target a missing link, such as whether a measured property persists after preparation or corresponds to an interface observation. These are proposed studies, not completed experiments. Their purpose is to turn a broad biological explanation into a sequence of observations capable of supporting or challenging it.
A treatment-version table can make the proposed contrast concrete before outcomes are examined. Each row would identify the documented source, preparation version, measurement basis, site, and eligible clinical episodes. A blank would distinguish an unrecorded feature from an observed absence. The table could show that a nominal source comparison is actually supported only within one preparation pathway. That finding would narrow the research question without making the dataset useless. It could also identify an additional sample or clinical setting that would contribute genuinely new support. A later outcome model should inherit the contrast justified by this table rather than redefine the exposure merely because another coding choice produces a cleaner result. This sequence keeps the interpretation connected to the implanted objects and their documented histories.
Source and processing are related features of a graft strategy, but they should remain distinguishable when a study seeks to explain outcomes. The biological question raised by the Cao review is most testable when the delivered construct, selection process, and outcome definition are preserved together. The framework developed here connects those requirements without estimating a new clinical effect. It allows a study to report the contrast its data actually contain, whether that is a source comparison or a complete-strategy comparison. Clear exposure definitions make both favorable and inconclusive findings more scientifically useful.