Signal-Amplification Strategies for Gold-Nanoparticle Lateral Flow Assays: A Structured Review of Limit-of-Detection Improvement Relative to Conventional GNP-LFA

1. Introduction

Point-of-care diagnostic platforms that can be deployed outside centralized laboratories are increasingly relevant to conditions where early detection depends on measuring biomarkers present at very low plasma concentrations, such as p-tau217, GFAP, NfL, and the Aβ42/40 ratio in Alzheimer’s disease. A GNP-LFA capable of resolving these biomarkers at clinically meaningful concentrations would need a limit of detection well below what conventional colorimetric GNP-LFA typically achieves, which is the motivation for surveying amplification strategies systematically here.

Gold-nanoparticle lateral flow assays are widely used because they are simple, rapid, and inexpensive, but their conventional colorimetric readout often has limited analytical sensitivity and poor quantification, which is a recognized bottleneck for applications such as early disease diagnosis [1]. This limitation is especially important when targets are present at low abundance in the sample, when matrix effects reduce visual detectability, or when a point-of-care setting requires a result that does not depend on laboratory-grade instrumentation to interpret. Accordingly, a substantial body of work has explored sensitization approaches intended to improve the limit of detection without abandoning the basic lateral-flow format.

The review question addressed here is: among signal-amplification and sensitivity-enhancement strategies applied to gold-nanoparticle lateral flow assays, which strategies produce the greatest improvement in limit of detection relative to a conventional gold-nanoparticle comparator, and by what magnitude? Eligibility was defined broadly enough to admit silver and gold enhancement, enzyme-linked and nanozyme-catalyzed amplification, dual-label and aggregation-based approaches, magnetic and electrokinetic pre-concentration, flow-control and pre-enrichment methods, nanoparticle architecture modification, and SERS-assisted readouts, provided the reported gain was measured against a conventional GNP-LFA baseline rather than a different assay format or label material (Section 2.2).

Two recent reviews cover overlapping ground. One surveys nanoparticle-based signal-amplification strategies for lateral flow immunoassays broadly, addressing label chemistry, membrane and conjugate-pad optimization, and multiplexing rather than a quantitative cross-mechanism comparison against a defined baseline [2]. Another takes a bibliometric approach to spherical-nanoparticle labels specifically, cataloguing preparation and optimization strategies across noble-metal, fluorescent, nanozyme, SERS, and magnetic labels, and notes that systematic reviews focused specifically on spherical-nanoparticle-labeled LFIA for clinical detection remain scarce [3]. Neither review restricts its comparator to a conventional GNP-LFA baseline or reports fold-improvement in LOD as a common, extractable outcome measure across mechanism categories; the present review differs by applying that comparator restriction as an eligibility criterion from the outset (Section 2.2) and by extracting a single quantitative outcome, fold-improvement in LOD, wherever the source study made this determinable.

Twenty-nine studies met this criterion and contributed thirty-six comparator arms across nine mechanism categories, spanning infectious disease targets, cancer and organ-function biomarkers, and small-molecule residues in matrices including buffer, serum, plasma, saliva, and food or agricultural extracts (Section 3.2). This evidence base supports a comparison across most of the major amplification strategies described in the literature, though, as discussed in Section 4.1, the number of independent studies per category still varies considerably, and some categories (SERS-assisted readout, electrokinetic pre-concentration) remain represented by a single study each.

2. Methods

2.1. Review Design and Scope

This review examines signal-amplification and sensitivity-enhancement strategies applied to gold-nanoparticle lateral flow assays (GNP-LFA/AuNP-LFA), evaluating each strategy against its own conventional (unenhanced) GNP-LFA comparator wherever a comparator was reported.

Screening and eligibility decisions were made by a single reviewer, without a second reviewer independently screening the same records and reconciling disagreements. This review is therefore a structured literature review rather than a systematic review conducted to full PRISMA dual-reviewer standard. Section 2.7 discusses what this means for how the findings should be read.

2.2. Eligibility Criteria

Records were eligible if they:

• Described a GNP/AuNP-based lateral flow assay (colorimetric, fluorescent, or SERS-based readout) incorporating some form of signal amplification, sensitivity enhancement, or architectural modification relative to a conventional spherical-AuNP LFA; and
• Reported a limit of detection (LOD) for the enhanced assay together with a stated comparator LOD, either (a) from a same-study unenhanced or conventionally labelled arm, or (b) from a previously published conventional GNP-LFA for the same or a comparable analyte.

The second criterion admits both same-study and literature-baseline comparators because many amplification papers report their enhanced LOD against a literature baseline rather than running a parallel unenhanced arm in-house. A same-study-only requirement would exclude a large share of the relevant literature for reasons unrelated to the quality of the evidence. A literature-baseline comparator was accepted only where the baseline was explicitly stated by the study’s authors, rather than inferred by us, and only where the comparator itself was a conventional GNP-LFA rather than a different assay format (e.g., ELISA) or a different nanoparticle label material entirely (e.g., quantum dots, latex, or carbon nanoparticles used as the primary label rather than gold). For same-particle nanozyme comparisons, where the amplification mechanism under evaluation is the catalytic or enzymatic step itself, the “unenhanced” arm is the identical nanoparticle construct with that catalytic step omitted, rather than a plain unmodified spherical AuNP; the particle architecture is held constant so that the reported gain isolates the effect of the catalytic amplification step being evaluated.

Records were excluded if they:

• Reported a GNP-LFA method with no unenhanced comparator of any kind (development-only or kit-description papers);
• Compared a GNP-LFA against a different label material or a different assay format rather than a conventional AuNP-LFA baseline;
• Were themselves review articles rather than primary research;
• Were preprints without a located peer-reviewed counterpart at the time of screening; or
• Were represented only by a secondary source (e.g., a commercial vendor’s summary page) rather than the primary peer-reviewed article.

2.3. Search Strategy

Candidate records were identified through a set of primary discovery searches, each covering a distinct mechanism category, combined with “gold nanoparticle” and “lateral flow.” Table A1 (Appendix A) logs the primary discovery queries as run, in the order executed, together with the approximate number of distinct candidate records each query contributed to the tracking table before deduplication against records already logged. Additional targeted searches (not logged in Table A1) were subsequently run per-record to confirm exact author lists, journal details, DOIs, and to locate peer-reviewed versions of two records initially found as preprints.

Search results were drawn from indexing and full-text sources spanning PubMed/MEDLINE-indexed literature and publisher repositories (American Chemical Society, Elsevier/ScienceDirect, Wiley, Springer Nature, the Royal Society of Chemistry, and MDPI), along with preprint servers (arXiv, ChemRxiv) where preliminary versions of eventually-published work were encountered.

This approach differs from a single Boolean query executed against named databases (e.g., a fixed PubMed/Scopus/Web of Science search string with a documented date-stamped export), which is the more conventional systematic-review search method. The present search is a comprehensive, investigator-directed literature search rather than a database-exported systematic search.

Records were pulled from a single search session and compiled directly into a single tracking table as they were identified, so no separate deduplication step was required. A formal multi-database search conducted in named database interfaces would need an explicit deduplication step, and the resulting duplicate count should not be assumed to be zero.

2.4. Study Selection

Thirty-six candidate records were assessed at the full-text level against the eligibility criteria in Section 2.2. Seven were excluded: two for lacking any unenhanced comparator, one for comparing against a non-GNP-LFA baseline or different label material, three for being review articles rather than primary research, and one for being represented only by a secondary vendor summary rather than a locatable primary source. Two further candidate records were initially set aside as unreviewed preprints; both have since appeared in peer-reviewed journals (Analytical Chemistry and ACS Applied Materials & Interfaces respectively) and are included here citing the published version rather than the preprint.

Twenty-nine records met eligibility and were carried forward to data extraction (Section 2.5); the flow of records is summarized in Figure 1.

One included record, a dual-enhancement carcinoembryonic antigen assay, was reclassified during full-text reading. Its no-enhancement condition failed to produce any visually detectable signal at all, rather than a weaker-but-quantifiable one, so no numeric fold-change could be computed from it, and the paper’s own reported comparison was instead against different assay platforms entirely (fluorescent and magnetic-particle-based), which does not meet this review’s comparator criterion. The study is retained for its methodological description but its data are not entered into the quantitative fold-change synthesis. This distinction was identified through full-text reading rather than abstract-level screening.

2.5. Data Extraction

For each included record, the following were extracted independently for each reported comparator arm (a single paper could contribute more than one arm where multiple enhancement mechanisms or multiple analytes were reported): analyte, amplification/enhancement mechanism, comparator type (same-study or literature-baseline), baseline LOD, enhanced LOD, units, fold-improvement (calculated only where both baseline and enhanced LOD were stated in the same units by the study’s authors, not inferred or estimated by us), assay time where reported, and sample matrix. Where a study reported a magnitude for a related but distinct quantity, for example an increase in enzymatic catalytic activity relative to a reference enzyme rather than an increase in assay LOD relative to a conventional GNP-LFA, that magnitude was recorded but not treated as an LOD fold-improvement, to avoid conflating catalytic-activity gains with assay-sensitivity gains.

2.6. Risk-of-Bias Appraisal Approach

A full QUADAS-2 application was not feasible from the accessible full text alone, because domains such as blinding of LOD interpretation, replicate counts, and sample selection procedure were not consistently reported or extractable across the included studies. A partial, comparator-focused adaptation was applied instead, judging four domains per included study: sample selection, index test conduct, comparator determination, and flow and timing. The comparator-determination and flow-and-timing domains were judged directly from the tracked comparator type (Section 2.5): a same-study comparator, measured under matched reagents, protocol, and timing in the same paper, was judged low risk on both domains; a literature-baseline comparator, drawn from a different paper and potentially a different era, reagent batch, or protocol, was judged high risk on both domains, since differences between the two could reflect factors other than the amplification mechanism itself. The sample-selection and index-test-conduct domains are judged unclear throughout, because the methodological detail needed to judge them (blinding, replicate counts, sample selection procedure) was not consistently available in the accessible text. This is a narrower appraisal than a full QUADAS-2 application and should be read as such.

2.7. Limitations of This Search and Extraction Process

• Screening and eligibility judgments were made by a single reviewer without independent duplicate screening, which is a departure from full PRISMA-standard systematic review conduct.
• The search was investigator-directed and keyword-iterative rather than a single fixed Boolean export from named databases with a documented search date and hit count per database.
• A partial, comparator-focused risk-of-bias appraisal was applied (Section 2.6, Section 3.6) rather than a full QUADAS-2 application; two of the four domains judged (sample selection, index test conduct) are rated unclear throughout because the underlying methodological detail was not consistently extractable from accessible full text.
• A small number of included studies (noted in the extraction table) report a magnitude for the mechanism category as a whole, drawn from a systematic comparison paper, rather than a study-specific fold-change, because the specific study’s own comparator number could not be located in accessible full text; this is flagged per record rather than silently substituted.
• Fold-improvements are reported as single point estimates because the primary studies themselves typically report a single LOD value rather than a distribution; no confidence intervals, standard deviations, or standard errors are available to accompany the fold-improvement figures in this review, which limits assessment of how reliable any individual reported gain is.
• This review has not been prospectively registered (e.g., PROSPERO). A formal risk-of-bias appraisal, independent dual screening, and database-exported search are identified as priorities for a future, more rigorous version of this work rather than being addressed in the present version.

3. Results

3.1. Study Selection

Thirty-six candidate records were assessed at full-text level. Twenty-nine met eligibility and were carried into data extraction; seven were excluded (Figure 1; Section 2.4). The included records span nine mechanism categories: silver/gold enhancement, flow control, flow control/pre-enrichment, enzyme/catalytic (nanozyme) amplification, nanoparticle architecture modification, dual-label/aggregation, magnetic pre-concentration, electrokinetic pre-concentration, and SERS/optical hot-spot readout. One included record (a dual-enhancement carcinoembryonic antigen assay) contributed a mechanism description without a usable fold-change value, for the reason given in Section 2.4.

3.2. Study Characteristics

The 29 included records contributed 36 individual comparator arms, since several studies reported more than one enhancement mechanism or more than one analyte within the same paper. Analytes spanned infectious disease targets (SARS-CoV-2 nucleocapsid/RBD and IgG, Rift Valley fever virus nucleoprotein, E. coli O157:H7, Ralstonia solanacearum), cancer and organ-function biomarkers (carcinoembryonic antigen, prostate specific antigen, cardiac troponin I/T, ferritin, interleukin-12, pepsinogen I/II, transglutaminase 2, thyroid-stimulating hormone, human chorionic gonadotropin, hepatitis B surface antigen), and small-molecule residues (carbendazim, kanamycin, paraquat, aflatoxin B1). Sample matrices, where reported, included buffer, serum, plasma, saliva, and food/agricultural extracts. Table 1 summarizes the included studies by analyte, mechanism, and reported fold-improvement.

3.3. Fold-Improvement by Mechanism Category

Reported fold-improvements in LOD relative to a conventional GNP-LFA comparator, where a numeric value could be computed from author-stated baseline and enhanced LOD values, are summarized below by mechanism category and shown in Figure 2.

Silver/gold enhancement (7 arms, 4 independent studies): fold-improvements ranged from 3x (PSA, silver enhancement [7]) to 488x (SARS-CoV-2 RBD, gold enhancement [6]), with the same SARS-CoV-2 study reporting three enhancement variants within one comparator (silver enhancement 8x, galvanic Ag-to-Au replacement 61x, gold enhancement 488x [6]). A ferritin assay achieved 20x through on-device gold enhancement [13]. The lowest gains in this category (3-3.6x) came from silver enhancement alone (a salivary cortisol assay reported 3.6x [8], and a potato brown rot assay reported 33x via full gold enhancement [10]); the largest gains came from full gold-enhancement or galvanic-replacement protocols.

Enzyme/catalytic (nanozyme) amplification (5 quantifiable arms): fold-improvements ranged from 4x (kanamycin, gold/DNA-Cu2+ nanozyme [18]) to 1000x (E. coli O157:H7, avidin-biotin multi-enzyme construct [12]), with intermediate values of 40x (transglutaminase 2, chromium nitride nanozyme [30]), 100x (PSA, Au@Pt4L nanozyme [28]), and 200x (pepsinogen I/II, Pd@Ir bimetallic nanozyme [29]). This was the mechanism category with both the single highest confirmed fold-improvement (1000x) and the widest internal spread. Two further nanozyme studies (Au-Ir core-shell [16]; Au@Pt ultrathin-shell [17]) could not contribute a quantifiable LOD fold-change: one lacked a locatable baseline comparator in accessible text, and the other reported a 1063x figure that describes peroxidase-mimicking catalytic activity relative to horseradish peroxidase, not assay LOD relative to a conventional GNP-LFA, and is therefore not reported here as a sensitivity gain.

Nanoparticle architecture modification (5 arms): gold nanoshells gave a 26x improvement (TSH [19]); gold nanopinecones gave a 2x improvement against a same-study commercial spherical-AuNP comparator and a 24x improvement against a previously published GNP-LFA for the same analyte (CEA [21]); colloidal gold superparticles gave 39.79x (hCG) and 13.8x (HBsAg) against a traditional AuNP-LFA [27]. The two-comparator CEA result illustrates that the size of a reported fold-improvement depends heavily on which baseline is chosen: the same architecture change looks far more modest against a same-study spherical-AuNP arm than against an older literature baseline.

Dual-label and aggregation-based approaches (3 arms): a biotin-streptavidin dual-gold-nanoparticle network gave a 3x improvement for cardiac troponin I/T against literature-reported conventional systems [14]. This is the least specifically identified literature comparator in this review: the source paper itself describes the baseline only as “more commonly used systems,” without naming a specific method, so the 3x figure should be read with that inherent limitation in mind; it reflects a gap in the original study’s own reporting rather than a gap in this review’s extraction. Two fluorescent/colorimetric dual-mode devices, one using a gold nanoparticle-quantum dot hybrid label (SARS-CoV-2 [22]) and one using metal-enhanced fluorescence with a Cy3-gold nanoparticle conjugate (SARS-CoV-2 IgG [32]), each gave a 10x and 1000x improvement respectively when the fluorescent readout of the same device was compared with its own colorimetric readout.

Magnetic pre-concentration (3 arms): a magnetic bead-AuNP construct with gold nucleation gave a 10x improvement (cardiac troponin I [24]); magnetite nanoparticle pre-concentration gave a 5x improvement in coloration intensity for paraquat, with up to a further 333x sample-concentration factor achievable independent of the coloration gain itself [25]. A magnetic-gold nanostar hybrid combining magnetic pre-concentration with plasmonic nanoparticle architecture gave a 1000x improvement for Rift Valley fever virus nucleoprotein, the largest gain reported for any hybrid mechanism in this review [26].

Electrokinetic pre-concentration (1 study, 2 arms): isotachophoresis gave a 400x improvement at a 90-second assay time and a 160x improvement at a 5-minute assay time for a model IgG target [23], illustrating a direct trade-off between achievable sensitivity gain and assay duration within a single mechanism.

Flow control and pre-enrichment (6 arms across 2 studies): deceleration-membrane flow control gave improvements of 1.5x, 3.5x, and 7.5x across three membrane variants for carbendazim [5], and a dual independent sample inlet gave a 5x improvement for ferritin [13]. Test-zone pre-enrichment gave a 10-100x improvement for miR-210 mimic and a 10x improvement for human chorionic gonadotropin, but produced no measurable improvement for a third analyte tested in the same study (aflatoxin B1) [31].

SERS/optical hot-spot readout (1 study): a dual-flow SERS device achieved a limit of detection below 0.5 uIU/mL for thyroid-stimulating hormone [4]; the source study did not report a numeric baseline comparator, so no fold-change is given for this arm.

3.4. Mechanisms Reporting No Meaningful Improvement

Two arms in this review reported no meaningful sensitivity gain over a conventional or visual-readout comparator. A photothermal speckle-imaging readout and a colorimetric-plus-machine-learning readout, both applied to the same commercial Salmonella LFA, each performed comparably to standard visual interpretation rather than improving on it [11]. Separately, test-zone pre-enrichment, which gave a 10-100x improvement for miR-210 mimic and a 10x improvement for human chorionic gonadotropin in the same study, produced no improvement for aflatoxin B1 [31]. These results are reported alongside the positive findings above because a synthesis limited to positive results would misrepresent how consistently these mechanisms perform across different analytes and readout formats.

3.5. Quantitative Synthesis by Mechanism Category

Table 2 summarizes the median, minimum, and maximum reported fold-improvement per mechanism category, computed directly from the confirmed numeric values in the extraction table. This is a descriptive summary, not an inferential meta-analysis: no pooling, weighting by study size, or confidence intervals were computed, given the heterogeneity of analytes, matrices, and comparator types across studies (Section 4.4).

3.6. Risk-of-Bias Appraisal

Table 3 presents the per-study risk-of-bias judgments described in Section 2.6. Twenty-four of the 29 included studies were judged overall low risk, reflecting a same-study comparator measured under matched conditions. One study [14] was judged high risk, since its reported comparator was a literature-referenced set of “more commonly used systems” rather than a same-study arm. One study [21] was judged mixed, since it reported one same-study arm and one literature-baseline arm with different risk profiles. Three studies [4,9,15] were judged unclear or not applicable at the comparator-determination domain, for the reasons given in Section 3.3 and Section 3.4. Two studies [16,17] were not applicable, since neither contributed a usable LOD fold-change to the synthesis. The sample-selection and index-test-conduct domains are rated unclear across all studies, for the reason given in Section 2.6, and this should be read as a limitation of the appraisal itself rather than a finding that these aspects of the underlying studies were necessarily deficient.

4. Discussion

4.1. Cross-Category Comparison

The mechanism categories varied substantially in both the magnitude and the consistency of reported sensitivity gains. Enzyme/catalytic (nanozyme) amplification spanned the widest range of any category (4x-1000x across 5 quantifiable arms [12,18,28,29,30]) and included the single largest confirmed same-mechanism gain reported in this review (see also [33] for related catalytic nanozyme labels, not among the included records). Hybrid approaches combining magnetic pre-concentration with plasmonic nanoparticle architecture [26], and electrokinetic pre-concentration on its own [23], each reached 400x or higher under at least one tested condition. Silver/gold enhancement was the most consistently represented category, appearing across four independent studies [6,7,8,10,13], but produced a comparatively modest range (3x-488x) with most values concentrated below 40x.

This comparison should be read with several caveats rather than as a ranking of mechanisms by inherent effectiveness. Most categories in this review are supported by a small number of studies, as few as one for SERS/optical hot-spot readout and for electrokinetic pre-concentration. This single-study representation is consistent with the broader signal-amplification literature: a prior review focused specifically on SERS, photothermal, and photoacoustic plasmonic readouts for LFA likewise treats these as an actively developing but comparatively less-populated area relative to enzyme/catalytic and metal-enhancement approaches [1]. A single unusually high or low result can therefore shift the apparent range for an entire category. The studies included here were also identified through a keyword-driven search rather than an exhaustive database export (Section 2.3). This may under-represent mechanisms that are described using terminology not covered by the search terms used. Studies reporting large fold-improvements may also be more likely to be published, and more likely to surface in a literature search, than studies reporting modest or null results. The ranges reported here therefore likely skew toward the more favourable end of what any given mechanism can achieve in practice. Section 3.4’s two null findings, both occurring within studies that reported strong gains for other analytes in the same paper, are a reminder that a mechanism’s reported ceiling does not transfer automatically across analytes, matrices, or assay formats.

4.2. What the Comparator Type Tells Us

Several included arms illustrate that a reported fold-improvement is only meaningful in light of what it was measured against. The carcinoembryonic antigen nanopinecone study [21] (Section 3.3) reported a 2x gain against a same-study spherical-AuNP comparator and a 24x gain against an older literature-reported GNP-LFA for the same analyte. Both numbers are accurate, but they answer different questions. The 2x figure describes the incremental value of the architecture change alone, holding everything else constant. The 24x figure describes cumulative progress against the field’s historical baseline, which reflects many other changes beyond this one design choice. Reviews and proposals that quote only the larger of two such numbers risk overstating what a specific design choice actually contributes on its own. Where this review reports a literature-baseline comparator rather than a same-study one (noted per arm in the Extraction Table), the fold-improvement should be read as a statement about the field’s progress over time rather than a controlled, same-study effect size.

4.3. Practical Implications for GNP-LFA Assay Design

Taken together, the included studies suggest that the largest sensitivity gains in this literature come from mechanisms that alter the detection chemistry itself, such as catalytic/enzymatic signal deposition, metal enhancement, or fluorescence. Mechanisms that only alter how the label is loaded onto the particle, or how the strip is physically constructed, tended to produce smaller gains. Flow-control approaches (deceleration membranes, dual sample inlets, test-zone pre-enrichment) gave more modest and more analyte-dependent gains, and in one case no gain at all. This suggests they may be better suited as a complementary refinement to an already-sensitive assay rather than a primary sensitivity strategy on their own. Approaches combining two mechanisms in the same device, such as magnetic pre-concentration paired with plasmonic nanoparticle architecture [26], produced some of the largest gains reported here. This may reflect that pre-concentration and signal amplification address different limiting steps: getting enough target to the detection zone, versus generating enough signal per bound target. The two can therefore be complementary rather than redundant.

4.4. Limitations

The limitations of the search and extraction process itself are described in Section 2.7. A descriptive quantitative synthesis (median, minimum, and maximum per category) is provided in Section 3.5, and a partial, comparator-focused risk-of-bias appraisal is provided in Section 3.6; neither substitutes for formal inferential meta-analysis or a full QUADAS-2 application. The cross-category comparison in this section remains descriptive rather than statistical: no formal meta-analytic pooling (e.g., a random-effects model with computed confidence intervals across studies) has been attempted, given the heterogeneity of analytes, matrices, and comparator types involved, and given that several categories are supported by very few studies. A meta-analytic approach may be appropriate for a subset of these mechanisms (for example, silver/gold enhancement, which has the largest number of independent studies) if a future version of this review focuses more narrowly on a single mechanism category with harmonized comparator definitions and obtains variance data from the primary studies or their authors.

5. Conclusion

Across 29 included studies and 36 comparator arms, GNP-LFA amplification strategies produced fold-improvements in limit of detection ranging from no measurable gain to 1000x, with the size of the gain depending on both the mechanism and the analyte rather than following a single predictable curve. Strategies that intervene directly in the detection chemistry, catalytic signal deposition, metal enhancement, and fluorescence, produced the largest and most frequent gains in this evidence base, while strategies that only alter flow dynamics or label loading produced smaller and more analyte-dependent gains, including two documented cases of no improvement at all. Combining a pre-concentration step with a signal-amplification step in the same device produced some of the largest gains reported here, suggesting these two categories of intervention address different limiting steps in the assay rather than competing for the same source of sensitivity.

For a GNP-LFA under development, this evidence base points toward catalytic and metal-enhancement strategies as the strategies with the largest and most frequently reported gains, with pre-concentration as a plausible complementary addition rather than a substitute. This should be read as a preliminary signal rather than a settled recommendation: most mechanism categories, including these two, are supported by only a handful of independent studies, and the null results documented here show that a strategy’s gain in one study does not guarantee a gain for a different analyte or matrix. Any specific mechanism chosen on the basis of this review should be validated directly on the target analyte before being relied upon.

This review’s search was investigator-directed and screened by a single reviewer rather than conducted as a database-exported, dual-reviewer systematic search, and only a partial, comparator-focused risk-of-bias appraisal has been applied rather than a full QUADAS-2 assessment (Section 2.6, Section 2.7, Section 3.6 and Section 4.4).