Acute PVD: Twenty-Six Papers, Assembled

This page is the reading behind poster F.127. It lists the twenty-six sources the poster draws on, grouped by the question each one answers, and gives for every entry the design, the cohort, the length of follow-up, the figure it supports, and one limitation its own authors declared.

It is written for clinicians who want to check a number rather than accept it. Nothing here is a recommendation. Where two papers disagree, both are listed and the disagreement is described rather than resolved.

The triage protocol itself is being prepared for journal submission and is not reproduced here.

Castilla Marti M. Risk Stratification in Acute Posterior Vitreous Detachment: A Structured Evidence-Based Triage Protocol. Poster F.127, 29th EVER Congress, Florence, 16 October 2026.

Figures are quoted as each paper prints them. Where a paper prints a figure twice with different values, both are given and the discrepancy is recorded in section 7. Editorial comment appears only in the methodological note that closes section 1 and in section 7.

1. How common is a retinal break, and why do the studies disagree

The rate of a concurrent retinal break in an acute symptomatic posterior vitreous detachment (PVD) is reported anywhere from 8.2% to 47.6%, and the most-quoted pooled figure across ten cohorts is 21.7%, a figure that does not reproduce from its own table (see 7.4). That spread is close to tenfold, and almost all of it is catchment rather than biology. A retina-only referral practice, a community clinic seeing self-referred floaters, and a national registry are counting different populations under one name. The entries below state the setting before they state the rate.

The delayed rate has a second problem on top of the first, which is the subject of the note that closes this section.

Coffee RE, Westfall AC, Davis GH, Mieler WF, Holz ER. Symptomatic posterior vitreous detachment and the incidence of delayed retinal breaks: case series and meta-analysis. Am J Ophthalmol. 2007;144(3):409-413. doi:10.1016/j.ajo.2007.05.002
Retrospective case-control series with a meta-analysis of ten studies. Own series 219 patients, of whom 201 had no break at presentation and 136 were re-examined within four weeks; pooled set 1,568 patients. Mean follow-up of the 136 approximately 1.6 years.
Prints a pooled concurrent rate of 21.7% and the range "from 47.6% to 8.2%", neither of which reproduces from its own table (see 7.4); a pooled delayed break rate of 1.8% (range 0.9% to 3.2%); its own delayed rate of 1.5%, two of 136; an odds ratio of 9.3 (95% CI 6.6 to 13) for pigmented vitreous cells, vitreous haemorrhage or retinal haemorrhage against a break at the initial examination, and 15.6 (7.1 to 34.5) against a delayed break; and that 24 of the 29 delayed breaks, 82.8%, had one of those signs or new symptoms.
Declared limitation: "The retrospective design of some of the studies included in the meta-analysis is a limitation of this research, because it is difficult to determine quality of examinations and data recording retrospectively."
Internal discrepancies, see 7.4.

Jindachomthong KK, Cabral H, Subramanian ML, Ness S, Siegel NH, Chhablani J, et al. Incidence and risk factors for delayed retinal tears after an acute, symptomatic posterior vitreous detachment. Ophthalmol Retina. 2023;7(4):318-324. doi:10.1016/j.oret.2022.10.012
Retrospective, consecutive, observational case series. 389 eyes of 389 patients with an acute symptomatic PVD and no concurrent tear or retinal detachment, drawn from 446 eyes of which 39, 8.7%, had a concurrent tear. Mean follow-up 666 days, median 564 days.
Supports a Kaplan-Meier failure estimate of 7.39% for a delayed tear by 6.24 years, on 20 raw events; a median time to the delayed tear of 44 days; that only 45% of the delayed tears had occurred by six weeks and the remaining 55% after, accounting for 2.83% of the whole cohort; that 25% of the eyes with a delayed tear had new or increased symptoms when it was found; and univariate hazard ratios of 6.48 (2.65 to 15.88) for lattice degeneration, 3.25 (1.06 to 9.95) for myopia and 2.55 (1.04 to 6.25) for age under 60, of which only lattice survived multivariate adjustment at 4.91 (1.66 to 14.53). The pseudophakia hazard ratio of 0.16 rests on one event in 99 eyes.
Declared limitation: "Like all retrospective studies using charts from multiple clinicians, it is impossible to definitively separate our missed retinal tears from delayed retinal tears."
Internal discrepancies, see 7.4. Table 1 of this paper is the subject of the methodological note below.

Richardson PS, Benson MT, Kirkby GR. The posterior vitreous detachment clinic: do new retinal breaks develop in the six weeks following an isolated symptomatic posterior vitreous detachment? Eye (Lond). 1999;13(Pt 2):237-240. doi:10.1038/eye.1999.58
Prospective clinic series, single vitreoretinal fellow examining at both visits. 107 patients referred, 105 with a confirmed PVD, 96 given a second appointment five weeks later.
Supports nine patients, 8.4%, referred for laser after the first visit, comprising six round holes, two horseshoe tears and one area of lattice with holes; two patients, 2.1%, found to have pre-equatorial round holes at the second visit, in both of whom a vitreous haemorrhage had obscured the first-visit view; and the statement that no patient in whom a full examination was possible at the first visit developed further retinal pathology.
Declared limitation, no sentence uses the word: "It was not possible to state from this study whether breaks developed between presentation to casualty and the first PVD clinic a few days later, nor can it be stated whether the retinal breaks in patients with vitreous haemorrhage developed between the PVD clinic visits."
Internal discrepancies, see 7.4.

Bond-Taylor M, Jakobsson G, Zetterberg M. Posterior vitreous detachment - prevalence of and risk factors for retinal tears. Clin Ophthalmol. 2017;11:1689-1695. doi:10.2147/OPTH.S143898
Retrospective case-control analysis of medical records, tertiary eye centre, Sahlgrenska University Hospital. 365 patients presenting with PVD-related symptoms for the prevalence figures; a larger case-control set of 426 for the risk-factor comparisons, 320 without and 106 with a tear. Follow-up 4.5 years.
Supports a concurrent tear rate of 14.5%, 53 of 365, and a vitreous or retinal haemorrhage rate of 22.7%; visual impairment (P=0.024), haemorrhage at examination (P<0.001) and symptom duration under 24 hours (P=0.004) as predictors, with flashes not constituting an extra risk at P=0.135; and a risk ratio of 17.7 (95% CI 2.2 to 145) for a retinal detachment in the same eye over the follow-up in eyes that had a tear at presentation. No odds ratios or confidence intervals are printed for the predictors, only P values.
Declared limitation: "Instead, this study used diagnosis codes as the definition of disease, which is a limitation of the study."
Note on this paper's variable denominators: Table 2 prints a different n for almost every row. The Weiss-ring row rests on 19 tear eyes, not 106.

Hollands H, Johnson D, Brox AC, Almeida D, Simel DL, Sharma S. Acute-onset floaters and flashes: is this patient at risk for retinal detachment? JAMA. 2009;302(20):2243-2249. doi:10.1001/jama.2009.1714
Systematic review and meta-analysis, seventeen studies, part of the Rational Clinical Examination series. No single pooled patient total is printed. Delayed-break analysis anchored at six weeks.
Supports a summary prevalence of a retinal tear of 14% (95% CI 12% to 16%) in patients self-referred or referred to an ophthalmologist; a six-week delayed tear incidence of 3.4%; positive likelihood ratios of 5.0 (3.1 to 8.1) for subjective vision reduction, 10 (5.1 to 20) for vitreous haemorrhage and 44 (2.3 to 852) for vitreous pigment, with a negative likelihood ratio for vitreous pigment of 0.23 (0.12 to 0.43); post-test probabilities of 45% (34% to 57%) after subjective vision reduction and 62% (45% to 77%) after vitreous haemorrhage; and that the presence of both floaters and flashes rather than one alone is not diagnostically useful, likelihood ratio 1.2 (1.0 to 1.3).
Declared limitation: "There are several limitations of our meta-analysis. First, data from all studies were obtained from ophthalmology clinics where patients were diagnosed as having PVD."
Caution. In the vitreous-haemorrhage section of Table 1, the Tasman 1968 row prints a positive likelihood ratio of 9.1 (3.2-6), an interval whose upper bound lies below its point estimate, and the Jaffe 1968 row prints 16 (6.5-40 with the interval unclosed. Neither row should be reused from this table.

Gishti O, van den Nieuwenhof R, Verhoekx J, van Overdam K. Symptoms related to posterior vitreous detachment and the risk of developing retinal tears: a systematic review. Acta Ophthalmol. 2019;97(4):347-352. doi:10.1111/aos.14012
Systematic review, thirteen studies, 2,651 participating eyes, mean age 62.2 years. Window for the delayed figures four to eight weeks after the first visit.
Supports an odds ratio of 19.8 (P=0.032) for more than ten floaters or a cloud in the vision against a retinal tear, pooled by Fisher's test across three studies; and pooled figures for isolated flashes, isolated floaters, both together, and haemorrhage or pigment.
Declared limitation: "Several limitations apply to this study. First, not all studies included did have detailed information on PVD complaints, such as the number of floaters, the intensity of floaters or presence of retinal and/or vitreous haemorrhage or blurring of vision."
Caution. This review's pooled estimator is a mean count of eyes per study with a confidence interval around that count, not a rate with a confidence interval. The accompanying percentages, 5.3%, 16.5%, 20.0%, 3.9% and 30.0%, are stated against a denominator the paper does not name, and where a denominator is printed they do not reproduce. The percentages from this paper should not be quoted as rates. The pooled odds ratio above is the figure that stands.

Kim SJ, Bailey ST, Kovach JL, Lim JI, Vemulakonda GA, Ying G, et al. Posterior Vitreous Detachment, Retinal Breaks, and Lattice Degeneration Preferred Practice Pattern. Ophthalmology. 2025;132(4):P163-P196. doi:10.1016/j.ophtha.2024.12.023
Consensus guidance, not a study. Approved by the Board of Trustees 13 September 2024 and published in the April 2025 issue. Christina J. Flaxel chaired the committee. The preceding edition is Ophthalmology 2020;127(1):P146-P181, which carries a published erratum at Ophthalmology 2020;127(9):1279; the two editions should not be cited interchangeably.
Supports the follow-up intervals in its Table 3: four to six weeks for a symptomatic PVD with no retinal break and no high-risk features; one to two weeks depending on severity where there is retinal haemorrhage, and weekly until resolved where there is vitreous haemorrhage, with ultrasonography as needed; and, separately, the recommendation that patients with any degree of vitreous pigment, vitreous or retinal haemorrhage, or visible vitreoretinal traction return for a second examination within six weeks. It states a concurrent break rate of 5.4% to 8% and a delayed rate of approximately 2%.
What the stratification uses. Table 3 stratifies on symptom status, presence and morphology of a break, lattice degeneration with or without holes, presence and type of haemorrhage, treated or untreated status, and whether the fellow eye has had a retinal detachment. Lattice degeneration and the fellow eye enter only the long-term surveillance rows (annually, or every six to twelve months); neither changes the interval for an acute symptomatic PVD. The term "high-risk features" is used in the table but defined only in the surrounding narrative. Age, sex, myopia, axial length, pseudophakia and family history appear in the risk-factor text and not in the follow-up table.
Declared limitation: none of its own work. "Adherence to these PPPs will not ensure a successful outcome in every situation."
Internal discrepancy, see 7.4.

Methodological note · the upper bound of the delayed-tear range

The range usually quoted for a delayed retinal tear after an uncomplicated acute posterior vitreous detachment runs from about 1.5% to about 8.7%. The upper bound does not survive a reading of the paper it is attributed to.

Jindachomthong et al tabulate Richardson et al at 8.7%, and derive it in the table footnote: "Richardson et al reported 107 patients who presented with concern for an acute, symptomatic PVD. However, 2 of these patients were found not to have a true PVD when reevaluated and 2 patients had a vitreous hemorrhage that obscured a prior retinal tear. Of the remaining 103 patients who were seen later in the PVD clinic, 9 patients developed a delayed retinal tear. The incidence of delayed retinal tears in this study was calculated as 9/103 = 8.7%."

Richardson reports those nine as a first-visit finding. The Results state: "Nine patients (8.4%) were referred for indirect laser photocoagulation following their first visit to the PVD clinic (Table 2)." The corresponding second-visit figure is two: "Two patients (2.1%) were found to have pre equatorial round holes at their second visit and were referred for indirect laser retinal photocoagulation (Table 2). In both patients a vitreous haemorrhage at the first visit had obscured the retinal view."

Richardson's Table 2 carries the same split. Of its eleven rows, nine record a break in the "First PVD visit" column and read "Same" at the second visit; two rows, patients 2 and 11, read "Vitreous haemorrhage" at the first visit and "Round hole" at the second. The nine of that first column is the nine of the Results sentence, and it is the same nine the abstract itemises as six round holes, two horseshoe tears and one area of lattice with holes.

The denominator of 103 appears nowhere in Richardson. The paper's stated denominators are 107 patients seen, 105 with a confirmed PVD, and 96 given second-visit appointments. Richardson's own percentage for the nine is 8.4%, which is nine of 107.

Two consequences follow. The first is that Richardson's conclusion runs the other way from the figure attributed to him: "If no retinal breaks are detected by thorough examination at presentation, further reviews, in the absence of increasing symptoms, may not be necessary." The second is that the two patients the footnote removes from the denominator, on the grounds that a vitreous haemorrhage had obscured a prior tear, appear to be the only two in that series with a genuinely delayed finding.

Two further details of the same table row are worth recording. It labels Richardson retrospective, where Richardson calls itself a prospective study. And it lists the symptomatic time period as six months, where Richardson's six months is the recruitment window and its mean symptom-to-presentation interval is five days.

The discrepancy was communicated to the corresponding author of Jindachomthong et al in September 2026.

Miguel Castilla Marti, September 2026.

2. What predicts a break at presentation

The findings in this section are not commensurable and are not ranked. Odds ratios, hazard ratios and likelihood ratios answer different questions, and an odds ratio from a multivariable triage model, a hazard ratio against a time-to-event endpoint and a likelihood ratio against a same-visit endpoint cannot be lined up by magnitude. Each entry states which estimator it reports and against which endpoint.

Three of the papers here print no effect estimate at all. Dayan 1996 and Natkunarajah 2003 contain no odds ratio, no confidence interval and no P value anywhere; van Overdam 2001 prints P values and predictive values only. Any effect estimate attributed to them would be an invention.

Read across the section rather than down it. The symptom the patient volunteers first is the one on which these cohorts agree least, and the findings that move the probability most are the ones that require a dilated pupil and an examiner willing to indent.

Karahan E, Karti O, Er D, Cam D, Aydın R, Zengin MO, et al. Risk factors for multiple retinal tears in patients with acute posterior vitreous detachment. Int Ophthalmol. 2018;38(1):257-263. doi:10.1007/s10792-017-0455-0 (published online 2017)
Prospective study of 376 consecutive patients presenting with floaters or flashes to a private clinic in Izmir, examined by one retinal specialist. Follow-up not stated.
Supports a concurrent tear rate of 14.4%, 54 patients with 71 tears; and univariate odds ratios against a tear at the initial examination of 26.096 (12.573 to 54.164) for tobacco dust, 4.608 (2.507 to 8.470) for retinal or vitreous haemorrhage, 3.002 (1.630 to 5.531) for multiple floaters and 2.05 (1.118 to 3.760) for flashing. Within the 54 tear eyes, haemorrhage raises the odds of multiple tears, 6.111 univariate and 7.085 (1.229 to 40.847) multivariate.
Declared limitation: "This study has some limitations. First, we did not determine whether PVD was complete or incomplete."
Internal discrepancies, see 7.4. Caution. The P values in Table 3 do not match their own Wald statistics (flashing: Wald 5.384, printed P=0.040; family history of detachment: Wald 3.288, printed P=0.020 and starred as significant, with a confidence interval of 0.982 to 10.816 that crosses 1). Quote the odds ratios and confidence intervals in preference to the P values.

Tanner V, Harle D, Tan J, Foote B, Williamson TH, Chignell AH. Acute posterior vitreous detachment: the predictive value of vitreous pigment and symptomatology. Br J Ophthalmol. 2000;84(11):1264-1268. doi:10.1136/bjo.84.11.1264
Prospective study, 200 consecutive phakic patients with a symptomatic PVD of under one month, London teaching hospital, examined by masked indented indirect ophthalmoscopy; plus a second prospective series of 115 patients presenting for primary retinal detachment repair. Follow-up not stated.
Supports a concurrent break rate of 12.5%, 25 of 200; Shafer's sign present in 23 of those 25, 92%, rising to a sensitivity of 95.8% and a specificity of 100% when one chronic lattice-associated hole is excluded; and the finding that symptomatology was not predictive, with flashes and floaters together carrying no significantly higher chance of a tear than floaters alone, χ²=1.65, P=0.19.
Declared limitation, no sentence uses the word: "Our series is also not a community based study and it is therefore difficult to completely exclude a possible referral bias in patients presenting to our tertiary referral practice, despite our attempts to encourage referral of all patients with acute PVD on an unselected basis."
Internal discrepancy, see 7.4.

Dayan MR, Jayamanne DGR, Andrews RM, Griffiths PG. Flashes and floaters as predictors of vitreoretinal pathology: is follow-up necessary for posterior vitreous detachment? Eye (Lond). 1996;10(Pt 4):456-458. doi:10.1038/eye.1996.100
Retrospective review of 295 consecutive eye-casualty presentations in 1993 and 1994. 169 patients given follow-up appointments, 157 attended. Per-patient follow-up duration not stated.
Supports a rate of a break or retinal detachment at presentation of 28.4%, 84 of 295, in a casualty population; the finding that 67% of the patients with a subjective reduction in vision had a tear or a detachment, against 23% with flashes and floaters, 18.9% with flashes and 18.5% with floaters alone; the sentence "We did not find isolated photopsia to be any more predictive of serious pathology than floaters"; and three new breaks found only at follow-up in patients with no further symptoms, 3.3% of all tears found and 1.9% of follow-up appointments.
Declared limitation, no sentence uses the word: "although our study was based on patients presenting to eye casualty to the exclusion of outpatient referrals, who tend to have a longer and less florid history of symptoms and who therefore may represent a different subgroup."
Internal discrepancies, see 7.4.

van Overdam KA, Bettink-Remeijer MW, Mulder PG, van Meurs JC. Symptoms predictive for the later development of retinal breaks. Arch Ophthalmol. 2001;119(10):1483-1486. doi:10.1001/archopht.119.10.1483
Prospective study, Rotterdam Eye Hospital casualty department, ten months. 280 patients seen, 30 excluded for a break at presentation, 250 analysed. Re-examination at six weeks, or at two weeks where a haemorrhage was found.
Supports a delayed break rate of 5.2%, 13 of 250; and the finding, by logistic regression with backward elimination, that "only the combination of flashes and more than 10 floaters (group C floaters) or a cloud or curtain (group D floaters) at the initial examination and an increase in the number of floaters after the initial examination were significantly (P<.001) related to the development of new retinal breaks", with predictive values of 75.0% and 99.6%.
Declared limitation, no sentence uses the word: "because this study concerned patients with an initial isolated PVD, several patients with breaks or detachments detected at the emergency unit were directly forwarded for treatment while their data were erroneously not processed for the study."
Caution. The Comment section restates the same model as a list of three contributing factors, which reads as though flashes were independent. The Results sentence with "only the combination" is the governing statement.

van Overdam KA, Bettink-Remeijer MW, Klaver CC, Mulder PG, Moll AC, van Meurs JC. Symptoms and findings predictive for the development of new retinal breaks. Arch Ophthalmol. 2005;123(4):479-484. doi:10.1001/archopht.123.4.479
Prospective study of 270 consecutive patients with a symptomatic isolated PVD, designed to validate the 2001 cohort. Re-examination at six weeks, and at two weeks where a haemorrhage was found.
Supports a delayed break rate of 3.7%, 10 of 270; three predictors surviving backward elimination, group C or D floaters, retinal or vitreous haemorrhage, and an increase in floaters after the initial examination, with log-odds weights of 1.7, 3.6 and 3.6; and, pooling this cohort with the 2001 one to 520 patients and 23 breaks, an area under the curve of 0.94, a sensitivity of 91.3% (72.0 to 98.9) and a specificity of 83.1% (79.8 to 86.4), with a negative predictive value of 99.5% (98.9 to 100). No new retinal breaks were found in the 170 patients with none of the three findings. Flashes did not survive: the P value is 0.70 in Table 1, and the authors write "The small sample size may also explain why flashes are a significant factor in our previous study and not in the present study."
Declared limitation, no sentence uses the word: "Therefore, our conclusions may only be valid when using our examination method."
Internal discrepancies, see 7.4. The age analysis in this paper is the subject of 7.3.

Seider MI, Conell C, Melles RB. Complications of acute posterior vitreous detachment. Ophthalmology. 2022;129(1):67-72. doi:10.1016/j.ophtha.2021.07.020
Retrospective cohort, Kaiser Permanente Northern California, calendar year 2018. Outcomes captured within the health system to 365 days.
Its four denominators, which are not interchangeable. 8,305 is the full cohort and carries the triage and examination models. 7,362 is the refraction subset and carries the myopia odds ratios. 7,522 is the subgroup with an initially uncomplicated PVD and carries every late-event figure. 133 is the number of late events, and is printed as the N of the late-event regression table although the model is necessarily run over the 7,522 at risk.
Supports a concurrent tear rate of 5.4% and retinal detachment rate of 4.0%; triage odds ratios of 2.67 (2.18 to 3.29) for blurred vision, 2.11 for male sex, 1.76 for age under 60, 1.44 (1.18 to 1.76) for prior cataract surgery, 1.23 (1.05 to 1.44) for a symptom duration of one week or less, and 0.76 (0.65 to 0.88) for flashing symptoms; examination odds ratios of 56.98 for vitreous pigment, 6.01 for lattice and 5.85 for vitreous haemorrhage; a late-event rate of 1.8% in the uncomplicated subgroup at a median of 22 days; late-event odds ratios of 17.03 for vitreous haemorrhage, 8.03 for lattice and 5.28 for a history in the fellow eye; and the stratification of 12.4% against 0.7% by the presence of any of those three.
Declared limitation: "However, there are several notable limitations of the current study." The one that travels with the timing figure: "because follow-up after the initial examination varied substantially, our design was not well suited to pinning down the exact timing of late events, which may have occurred at a variety of intervals before the examination when they were diagnosed."
Caution. The variable is labelled prior cataract surgery, not pseudophakia; the paper never uses the word. The numerators behind 12.4% and 0.7% appear only inside Figure 2 and are not printed in the text.
Internal discrepancies, see 7.4.

Bond-Taylor 2017, Hollands 2009 and Gishti 2019 also carry predictor data and are entered in section 1.

3. The examination itself

Everything in section 2 assumes the examination found what was there. This section is about how often it does not: indentation against slit-lamp biomicroscopy, the Weiss ring as a marker, the Shafer sign, optical coherence tomography, and ultra-widefield imaging.

The through-line is asymmetry. The signs here confirm better than they exclude, and a clear anterior vitreous in an eye with a fresh PVD carries less weight than its reputation suggests.

Nixon TRW, Davie RL, Snead MP. Posterior vitreous detachment and retinal tear - a prospective study of community referrals. Eye (Lond). 2024;38(4):786-791. doi:10.1038/s41433-023-02779-3
Prospective observational case series, 1,010 patients recruited from community referrals, of whom 896 had a PVD at first assessment. Two-month follow-up with indented indirect ophthalmoscopy in every case; follow-up data available for 729 of 901 asked to attend.
Supports a concurrent rate of 8.8% of the whole cohort and 9.9% of eyes with a PVD; a delayed rate of 3% at two months; that 13 of 88 tear eyes, 15%, were identified only by indented indirect ophthalmoscopy and not by slit-lamp biomicroscopy, and three of 86 only by slit lamp; that a Weiss ring was absent in 26 of the tear eyes, 32%; and that Shafer's sign was present in 61 of the newly diagnosed tear eyes, 75%, and in 11 eyes, 1.3%, with a PVD and no break.
Declared limitation, no sentence uses the word: "it is likely some patients will have declined or been unable to attend for assessment and so the denominator for the entire study population is unknown."
Internal discrepancies, see 7.4.

Natkunarajah M, Goldsmith C, Goble R. Diagnostic effectiveness of noncontact slitlamp examination in the identification of retinal tears. Eye (Lond). 2003;17(5):607-609. doi:10.1038/sj.eye.6700456
Prospective masked comparison in 17 eyes of 17 patients already known to have a tear, giving 18 tears. A single masked consultant vitreoretinal surgeon re-examined every patient. Follow-up not stated.
Supports that 16 of 18 tears, 89%, were correctly identified at the slit lamp with a non-contact lens, and two, 11%, were missed, one very anterior and visible only on indentation at the ora serrata, one masked within pigmented lattice; and that no additional information was obtained from the Vitreofundus or three-mirror lens over the Superfield.
Declared limitation: "Our high pickup rates using the slitlamp may be due to an evaluation bias, as all the patients entered into the study were thought to have a retinal tear."
Caution. This paper enrolled only eyes already known to have a tear. It supports no break rate and no predictor, and the epidemiological figures in its introduction are quoted from other authors.
Internal discrepancy, see 7.4.

Ansari WH, Blackorby BL, Shah GK, Blinder KJ, Dang S. OCT assistance in identifying retinal breaks in symptomatic posterior vitreous detachments. Ophthalmic Surg Lasers Imaging Retina. 2020;51(11):628-632. doi:10.3928/23258160-20201104-04
Retrospective chart review, single institution, single physician, January 2017 to July 2018. Follow-up not stated.
Supports the use of posterior vitreous opacities on spectral-domain optical coherence tomography as a marker of a retinal break, with a sensitivity of 76.0% and a specificity of 75.5% in the tables and abstract, rising to 82.4% and 86.4% in eyes without a vitreous haemorrhage; and an interobserver Pearson coefficient of 0.77.
Declared limitation: "Limitations include the size of the patient database and reading only a single raster line."
Caution. This is the most internally inconsistent paper in the set. The abstract and tables report n=78 and a sensitivity of 76.0%; the Results report n=76 and a sensitivity of 65.2%, and cite the table for values the table does not contain. Further mismatches between text and Table 1 are noted in 7.4. The table values are the ones that reproduce from their own cells.

Oh JH, Oh J, Roh HC. Vitreous hyper-reflective dots in optical coherence tomography and retinal tear in patients with acute posterior vitreous detachment. Curr Eye Res. 2017;42(8):1179-1184. doi:10.1080/02713683.2017.1289226
Retrospective case series, 77 consecutive patients with an acute symptomatic PVD, March 2013 to February 2015. Mean follow-up interval 2.8 weeks in the tear eyes and 4.3 weeks in the others.
Supports a retinal tear rate of 27.3%; a positive likelihood ratio of 9.78 (3.02 to 31.63) for severe vitreous hyper-reflective dots; 6.93 (2.82 to 17.08) for vitreous haemorrhage; and, for vitreous pigment, a sensitivity of 23.8% and a specificity of 98.2%, the sensitivity resting on five of the 21 tear eyes.
Declared limitation: "This study has limitations, including its retrospective design."
Internal discrepancies, see 7.4.

Lin AC, Kalaw FGP, Schönbach EM, Song D, Koretz Z, Walker E, et al. The sensitivity of ultra-widefield fundus photography versus scleral depressed examination for detection of retinal horseshoe tears. Am J Ophthalmol. 2023;255:155-160. doi:10.1016/j.ajo.2023.07.010
Retrospective clinical validity and reliability study, Shiley Eye Institute, 2020 to 2022. 123 patients with 135 horseshoe tears already diagnosed on scleral depressed examination. Single timepoint, no follow-up.
Supports that 69 of 135 tears, 51.1%, were visible on foveal-centred ultra-widefield images and 66, 48.9%, were not; that sensitivity by quadrant was 17.1% superior, 32.0% inferior, 50.0% nasal and 85.5% temporal, P<0.001; that in nine patients with both image types, none of nine foveal-centred images and all nine de-centred images showed the tear; and the conclusion that ultra-widefield imaging alone is not sufficiently sensitive to exclude a horseshoe tear.
Declared limitation: "As all of the included study patients were known to have HSTs, the image reviewers were masked only to the HST location but not the presence of a HST itself. This lack of masking to the HST presence could influence the sensitivity of the graders in identifying HSTs, thereby possibly overestimating the true sensitivity beyond the reported values."
Caution. The numbers 69 and 66, with the percentages 51.1% and 48.9%, are printed in adjacent sentences for two different quantities: a right-eye and left-eye split, and the detection result. The coincidence is arithmetically possible and is a transcription hazard.

Tanner 2000 is the primary source for the Shafer sign in this set and is entered in section 2.

4. Vitreous haemorrhage and anticoagulation

Vitreous haemorrhage is the one feature on which the follow-up table of current guidance already stratifies an acute PVD, and it is also the feature most often quoted with a number the underlying paper does not contain. Two questions belong apart: whether anticoagulation raises the chance of a haemorrhage, and whether, once a haemorrhage is present, anticoagulation changes what lies behind it.

El-Sanhouri AA, Foster RE, Petersen MR, Hutchins RK, Miller DM, Evans TM, et al. Retinal tears after posterior vitreous detachment and vitreous hemorrhage in patients on systemic anticoagulants. Eye (Lond). 2011;25(8):1016-1019. doi:10.1038/eye.2011.106
Consecutive observational case series, 260 eyes of 260 patients, every one of them with a vitreous haemorrhage by inclusion criterion. Follow-up 0 to 1,077 days, mean 85, median 33; no patients lost.
Supports that 46% of the 123 patients taking an anticoagulant had a retinal tear against 72% of the 137 not taking one, relative risk 0.64; and 23% against 37% for a detachment, relative risk 0.62.
Declared limitation: "These limitations of our study are certainly typical for retrospective case series", referring to the sentence before it: "The presence or absence of other known risk factors for RT and RD (refraction, lattice, contralateral RD, family history) were non-uniformly captured in the charts and were therefore not analyzed or reported."
Caution, and it is the load-bearing one. This study has no group without a vitreous haemorrhage. The 2% to 4% figure sometimes paired with it to produce a multiple is quoted by these authors from the literature, not measured here, and no relative risk in this paper is published with a confidence interval.
Internal discrepancies, see 7.4.

Witmer MT, Cohen SM. Oral anticoagulation and the risk of vitreous hemorrhage and retinal tears in eyes with acute posterior vitreous detachment. Retina. 2013;33(3):621-626. doi:10.1097/IAE.0b013e3182671006
Retrospective chart review of consecutive patients from a single referral practice, 2005 to 2009; 336 eligible patients, one eye each. Average follow-up 10 months, range 2 to 60.
Supports the intra-study comparison this literature otherwise lacks: retinal tears in 54 of 118 eyes with a vitreous haemorrhage, 45.7%, against 59 of 218 without, 27.0%, P=0.0007; vitreous haemorrhage in 43% of patients taking aspirin, clopidogrel or warfarin against 31% not taking them, P=0.03; and, within the haemorrhage stratum, tears in 39% of those on those medications against 52% of those not, P=0.20.
Declared limitation, no sentence uses the word: "Analysis of eyes with acute PVD and VH, who presented with retinal detachment, was limited in this report because of the low number of these patients (4 eyes)."
Internal discrepancies, see 7.4.

Vangipuram 2023 and Uhr 2020 both carry vitreous-haemorrhage data and are entered in section 5. The AAO Preferred Practice Pattern, entered in section 1, is the source of the weekly-review recommendation.

5. Delayed pathology and its timing

The six-week convention rests on the belief that retinal breaks declare themselves inside six weeks. The cohorts here are the ones that looked past it, and the ones that recorded whether the patient noticed.

One caution belongs at the top. Uhr prints a figure for delayed breaks found beyond six weeks and a second figure for those found within it, and the two are easily transposed. Each entry states which side of the boundary its figure describes.

Uhr JH, Obeid A, Wibbelsman TD, Wu C, Levin H, Garrigan H, et al. Delayed retinal breaks and detachments after acute posterior vitreous detachment. Ophthalmology. 2020;127(4):516-522. doi:10.1016/j.ophtha.2019.10.020
Retrospective case-control study, 7,999 eyes with a PVD and a documented extended ophthalmoscopic examination at the first visit. Mean follow-up 252 days; 55.9% of eyes had more than six weeks of follow-up.
Supports a concurrent break rate of 16.0% and retinal detachment rate of 6.2%; delayed breaks in 209 eyes, 2.6% of all eyes and 14.0% of all breaks, at a median of 35 days; delayed detachments in 80 eyes, 1.0% of all eyes and 13.8% of all detachments, at a median of 75 days; that 116 delayed breaks, 55.5%, were found within six weeks and 93, 44.5%, after; that 26 delayed detachments, 32.5%, were found within six weeks and 54, 67.5%, after; that 40.2% of the eyes with a delayed break and 67.5% of those with a delayed detachment presented with new or worsening symptoms when it was found; and a hazard ratio for pseudophakia of 2.10 (1.27 to 3.50) against a delayed detachment and 1.09 (0.75 to 1.58) against a delayed break.
Declared limitation: "There are several limitations to our study." The substantive one: "Some of the delayed breaks in our study may have been present but eluded detection on initial evaluation, or some delayed RDs may have resulted from a previously missed break."
Internal discrepancies, see 7.4.

Vangipuram G, Li C, Li S, Liu L, Harrison LD, Lum F, et al. Timing of delayed retinal pathology in patients presenting with acute posterior vitreous detachment in the IRIS Registry (Intelligent Research in Sight). Ophthalmol Retina. 2023;7(8):713-720. doi:10.1016/j.oret.2023.04.004
Retrospective database study, IRIS Registry, July 2013 to December 2018. 434,046 eyes of 367,645 patients met criteria, from 12.8 million coded with a PVD. Outcomes over one year; a minimum of one year of follow-up was an inclusion criterion.
Supports that 10,518 eyes, 2.42%, were found to have a delayed or missed retinal break or detachment within the year; a median time to a break of 42 days and to a detachment of 51 days; hazard ratios of 9.30 (8.50 to 10.2) for vitreous haemorrhage, 3.91 for a history in the fellow eye, 2.61 for lattice and 1.42 for myopia; that 61,700 eyes, 14.2%, met high-risk criteria and 6.77% of them had a delayed event, at a median of 34 and 39 days; and that in the 657 eyes documented with vitreous haemorrhage alone, the median time to a break was 14 days.
Declared limitation: "There are several limitations to this study." The substantive ones: "The diagnosis of acute PVD was limited to ICD coding and was unable to be individually verified by chart review, OCT, or ultrasonographic imaging" and "retinal pathology discovered after the initial PVD may have been present at the first examination and was not noted by the examiner."
Internal discrepancies, see 7.4.

Patel PR, Minkowski J, Dajani O, Weber J, Boucher N, MacCumber MW. Analysis of posterior vitreous detachment and development of complications using a large database of retina specialists. Ophthalmol Retina. 2023;7(3):203-214. doi:10.1016/j.oret.2022.11.009
Multicentre retrospective observational study, Vestrum Health database, 65 practices and 316 physicians, 2015 to 2019. 9,635 eyes. Fixed six-month administrative window, binned at presentation, one month, one to three months and three to six months.
Supports an any-complication rate of 25.0%, isolated vitreous haemorrhage 13.1%, retinal break without detachment 16.0% and detachment 4.2%; that 8.0% of the isolated haemorrhages, 19.2% of the breaks and 25.8% of the detachments were first noted after presentation, and that 38.9% and 51.9% of those delayed breaks and detachments respectively occurred after the first month; mean times to a delayed event of 36.7, 36.4 and 44.9 days; and unadjusted odds ratios of 1.85 (1.13 to 3.04) for pseudophakia against a delayed detachment, 2.79 for lattice against any complication, 1.55 for male sex, and 7.37 for a detachment within six months where the fellow eye has had one.
Declared limitation: "Limitations of our study include its retrospective study design, in which we relied on accurate data reporting and medical coding on the part of participating physicians." And: "It is also not possible to determine whether a noted complication was new versus missed on prior examinations, potentially affecting the average times to complication."
Caution. Every odds ratio in this paper is univariable; there is no multivariable model. None should be described as adjusted.
Internal discrepancies, see 7.4. Caution. In Table 1, the extended-ophthalmoscopy row prints 6,166 eyes with complications against a row total of 5,395 and a cohort total of 2,409 eyes with complications, and the percentage column switches denominator between rows. Its Table 1 should not be reused without recalculation.

Jindachomthong 2023, Seider 2022, Richardson 1999, Coffee 2007 and Dayan 1996 all carry delayed-event data and are entered in sections 1 and 2.

6. Triage before examination, and advice after it

The two ends of the episode. Before the examination, the decision about who is seen today is made by whoever answers the telephone, working from reported symptoms alone. After the discharge, the safety net is a sentence spoken at the end of the consultation.

Three papers have tested either end. They are small, and two of them are explicit about what that costs them.

Balikov DA, Zhou Y, Miller JML. A telephone triage system for patients calling with symptoms of a posterior vitreous detachment. Ophthalmol Retina. 2023;7(6):516-526. doi:10.1016/j.oret.2023.01.003
Prospective cohort quality-improvement study at one academic department over four months in 2020. 211 patients screened, 193 included, all followed to a dilated examination within six weeks.
Supports eight retinal tears and six detachments, 14 events in all; a risk score derived from 7 of 23 questions and 15 of 70 possible answers; an area under the curve of 0.9205 for the multivariable model and 0.9049 for the score; that a score of 20 captured all 14 patients with a tear or a detachment while leaving out 84 of the 179 without one, 46.9% (Table 7 prints the complement, 53.07% of them included), and a score of 29 captured 13 of 14 while leaving out 72.6%; the final multivariable model's strongest term, an odds ratio of 12.01 (2.83 to 50.99) for constant blurred vision or a curtain or veil; and, for calling within 24 hours of symptom onset against more than 72 hours, an odds ratio of 3.01 (0.64 to 14.20) at P=0.1641, which is not significant.
Declared limitation: "Another limitation of our study is that it lacks independent validation." And: "The patient sample size of approximately 200 patients is significantly smaller than previously published big-data analyses of risk factors for symptomatic PVDs."
Caution. The paired percentages in the Results, such as 78.57% against 48.60%, are the prevalence of a risk factor within each outcome group, not the probability of a tear. The score's area under the curve of 0.9049 appears only inside a figure panel.

McCullagh D, Higham A, Best R. The BElfast Retinal Tear and detachment Score (BERT Score). Eye (Lond). 2021;35(5):1427-1430. doi:10.1038/s41433-020-1025-7 (published online 2020)
Derivation across 100 prospective and 53 retrospective presentations, then application to a separate cohort of 160. Cross-sectional at the presenting visit; no follow-up.
Supports a six-item score, male sex 1, myopia 1, blurred vision 2, shadow 2, vitreous haemorrhage 3, tobacco dust 4; a sensitivity of 90% (68.3 to 98.8) and a specificity of 80.7% (72.3 to 86.9) at a threshold of four or more, with a negative predictive value of 98.3% (93.9 to 99.8) and a positive predictive value of 40% (25.7 to 55.7); and tobacco dust present in 22 of the 24 tear or detachment eyes in the derivation audit.
Declared limitation, no sentence uses the word: "This was a real-life study, with recognisable flaws." And: "There was a small sample of patients, with only 24 in the RT and RD group." The authors add: "before this scoring system is more widely implemented, it needs to be validated."
Caution. The recommended threshold is printed two ways. The abstract and results use four or more; the discussion proposes referring everyone scoring three or more. Every reported sensitivity and specificity is computed at the four-or-more cut.
Internal discrepancies, see 7.4.

Singh AJ, Seemongal-Dass RR. The influence of counselling on patient return following uncomplicated posterior vitreous detachment. Eye (Lond). 2001;15(Pt 2):152-154. doi:10.1038/eye.2001.50
Retrospective analysis of casualty notes comparing 50 consecutive patients counselled by an ophthalmic nurse practitioner with a leaflet in 1998 against 50 historical controls from 1997, before the protocol existed. Casualty attendances compared over one year from presentation.
Supports that seven controls and six counselled patients returned within the year; that the tested quantity is the narrower one, returns with no change in symptoms, three of 50 against none of 50, Fisher exact one-tailed P=0.13, of which the authors write "This was not highly significant"; and that "No patients re-presented with retinal breaks, retinal detachments or vitreous haemorrhage."
Declared limitation: none. The paper has no limitations paragraph and the word does not appear in it.
Internal discrepancies, see 7.4.

7. Where the evidence contradicts itself

Three places where this literature does not agree with itself, and one inventory of papers that do not agree with themselves, set out so a reader can see the disagreement rather than inherit whichever side a review happened to quote.

7.1 Pseudophakia points in four directions

Uhr et al report a hazard ratio of 2.10 (1.27 to 3.50) for a delayed retinal detachment in pseudophakic eyes; in the same cohort, for a delayed tear, the hazard ratio is 1.09 (0.75 to 1.58) and not significant. Seider et al report an odds ratio of 1.87 (1.19 to 2.96) for a late event, and the variable in that table is labelled prior cataract surgery, which is a history available at triage and not an examination finding of pseudophakia. Patel et al report an odds ratio of 1.85 (1.13 to 3.04) for a delayed detachment in pseudophakic eyes, univariable and unadjusted. Jindachomthong et al report a hazard ratio of 0.16 (0.02 to 1.18), which rests on a single event in 99 eyes.

Four estimates, three endpoints, two exposure definitions, one of them unadjusted and one of them supported by one event. The direction of the association depends on what is being counted, and no pooled figure should be constructed from these four.

7.2 The flash behaves differently in seven cohorts from six groups

Karahan et al report an odds ratio of 2.05 (1.118 to 3.760), P=0.040, on univariate analysis. Tanner et al conclude that symptoms alone cannot identify which patients have a break, and that flashes with floaters carry no significantly higher chance than floaters alone. Bond-Taylor et al report that symptoms of flashes did not constitute an extra risk, P=0.135. Dayan et al find isolated photopsia no more predictive than floaters. Van Overdam et al find flashes predictive in 2001 only in combination with a heavy floater burden, and in 2005 flashes reach P=0.70, the authors noting that the small sample size of the earlier study may explain the difference. Seider et al report flashing symptoms at an odds ratio of 0.76 (0.65 to 0.88), P<0.001, and offer as an explanation that "patients with symptoms of significant vision loss (i.e., those at highest risk of RT) do not report what may be perceived as milder symptoms, leaving those who do report photopsias lacking more serious symptomatology."

One positive univariate estimate, three null results, one failed internal replication and one inverse association. A predictor that behaves like this across seven datasets is unstable rather than weak, and the instability is the finding.

7.3 Age and timing, on ten events

Van Overdam et al 2005 is cited for age as a risk factor. What the paper reports is this: "There was also no significant difference between the number of new retinal breaks in patients older than 63 years and patients aged 63 years and younger. However, all 5 patients with a retinal break at the 2-week follow-up visit were 63 years or younger, whereas all 5 patients with a retinal break at the 6-week follow-up visit were 63 years or older. The difference in the number of retinal breaks in patients older than 63 years and patients 63 years and younger was significant at both the 2-week (P= .006) and 6-week follow-up visit (P=.005)."

Those two P values compare counts of events between two age strata at a given visit, on ten events in 270 patients. They are not a test of age as a predictor of a break, and the authors say so: "Patient age was not found to be a significant predictive factor for the development of new retinal breaks." Age appears neither in that paper's risk-factor table nor in its final model.

Two larger cohorts do find a crude association between younger age and risk: Seider et al report an odds ratio of 1.76 for age under 60 in their triage model, and Jindachomthong et al a univariate hazard ratio of 2.55 (1.04 to 6.25) for age under 60, which does not survive adjustment for lattice degeneration.

The distinction matters because the two claims license different actions. A predictor changes who is followed. A modifier of timing changes when.

7.4 Papers that disagree with themselves

Declared here because each of these figures is quoted somewhere, and a reader checking one against another will find the mismatch without explanation if it is not stated. None of it is a reason to discard the paper concerned. All of it is a reason to cite the primary text rather than a summary of it.

Coffee 2007. The per-study patient counts in its table sum to 1,674 against a printed total of 1,568, the difference being exactly one row. The pooled rate of 21.7% does not reproduce from that table: 300 of 1,568 is 19.1%. The stated maximum of the range, 47.6%, does not reproduce from any row; the highest is 46.0%. And the three risk factors named in the Methods are not the three named in the Results and abstract.

Richardson 1999. The median interval from symptom onset to presentation in the break patients is printed as 6 days in one paragraph and 18 days in another, with the same range. The combined-symptom group is 31 patients in the text and 36 in Table 1. One table percentage, 19.0%, does not reproduce from its own numerator and denominator.

Jindachomthong 2023. The Discussion states that "w 40% of our patient cohort was found to have retinal tears after 1 year", where the Results give 63.46% of the twenty delayed tears occurring within a year, that is about 36.5% of the twenty. As printed, the Discussion asserts a cohort-level rate roughly sixty times the paper's own figure. The Discussion also reports myopia and age as not significantly associated without distinguishing the univariate results, in which both were.

Hollands 2009. Table 1 prints a confidence interval for one row whose upper bound lies below its own lower bound and point estimate, and another that is unclosed.

Gishti 2019. The pooled haemorrhage estimate is printed three ways across the abstract, the results and the figure, with two different means and two different confidence intervals. The flashes-and-floaters confidence interval is printed two ways. The percentages do not reproduce from any printed denominator. A study appearing in three of its forest plots has no entry in its reference list.

AAO Preferred Practice Pattern 2025. The proportion of patients with an initial break who develop additional breaks over long-term follow-up is printed as "5% and 14%" in two places and as "10% to 16%" in a third, in the section on treated breaks, with an overlapping citation.

Bond-Taylor 2017. The same 365-patient cohort is given a February-to-July accrual window in the abstract and a February-to-April one in the methods and results.

Nixon 2024. The abstract describes recruitment over two years; the Methods describe an eighteen-month period. And the three vision categories sum to 1,001 of 1,010 patients, with one of the three percentages not reproducing.

Karahan 2018. The photopsia counts in Table 2 sum to 52 in the column of eyes with a tear, against the 54 declared. The sex percentages in Table 1 are transposed between columns. Three variables the Results describe as not differing between groups are asterisked as significant in Table 1. And the Wald statistics in Table 3 are not consistent with the P values printed beside them.

Tanner 2000. Table 2 places all ten vitreous haemorrhages in the no-break column, while the Discussion describes a patient who presented with both a retinal tear and a mild vitreous haemorrhage.

Dayan 1996. The count of isolated PVDs is 189 in one place and 188 in another. The risk of a break in the flashes-alone group is 18.9% in the Results and 11.6% in the Discussion, and the 11.6% is the share of all breaks occurring in that group, transplanted into a within-group risk statement. The symptom-group counts sum to 86 against a stated 84 breaks, and the share-of-breaks percentages reproduce against 86.

van Overdam 2001. The new-break rate is 5.2% in the abstract and results and 5.3% in the comment; 13 of 250 is 5.2%.

van Overdam 2005. The floater-group counts in the text sum to one fewer than the cohort while the table's sum correctly. The number of break patients reporting reduced vision at the two-week visit is three in the text and four in the table.

Seider 2022. The retinal detachment rate is 4.0% in the abstract and results and 4.3% in its own comparison table, and the combined rate of 9.7% in the Discussion reproduces only from the latter. The two component percentages of the late-event sentence reproduce against the full cohort while the rate they sit beside reproduces against the uncomplicated subgroup.

Ansari 2020. The study size is 78 in the abstract and tables and 76 in the Results, and the two sets of diagnostic figures differ throughout: sensitivity 76.0% against 65.2%, and, in the haemorrhage-free subset, 82.4% and 86.4% against 63.8% and 93.7%, with the Results citing the table for values it does not contain. Fifteen further text-against-table mismatches run through its Table 1.

Oh 2017. The mean number of dots in the moderate category is printed as equal to the maximum of its own range.

Witmer 2013. The P value for vitreous haemorrhage by anticoagulant use is printed as 0.03, 0.034 and 0.0337 in three places, and two reported findings do not meet the Bonferroni threshold the authors set themselves.

El-Sanhouri 2011. The abstract reverses the two treatment arms relative to the Results and Table 1. The P value for the tear comparison is 0.0002 in the abstract and 0.00002 in the Results and table. The design is called a consecutive observational case series in the Methods and a retrospective case series in the Discussion. The drug subgroups sum to more patients than the anticoagulated group contains.

Vangipuram 2023. The break and detachment counts sum to 896 more than the stated total of delayed events, with no statement that eyes with both are counted twice. The median time to a delayed detachment in the non-specialist group is 55 days in the table and 56 in the text.

Uhr 2020. The delayed-break rate in eyes with no risk factor is 0.7% in the Discussion and 0.8% in the Conclusion; 66 of 7,999 is 0.8%. The age effect is labelled an odds ratio in the abstract and a hazard ratio in the table.

Patel 2023. The male and female counts in Table 1 sum to twelve fewer than the cohort, while the risk-factor table's sum correctly. The count of concurrent detachments among eyes presenting with a vitreous haemorrhage is 101 in the text and 107 in the table, and only 107 reproduces against the paper's own derived figures. One confidence interval has transposed digits between text and table.

Balikov 2023. A delayed-event rate labelled "of all patients" is computed on a subgroup. A figure legend cross-references a table the manuscript does not contain. And the published score runs against its own model on one item: diabetes carries an odds ratio of 6.15 (1.01 to 37.55) in Table 4, but the score in Table 6 gives 1 point to diabetic patients and 5 to non-diabetic ones. The authors state that points were set approximately in proportion to the odds ratios, with small adjustments; this one is not explained.

McCullagh 2021. Specificity is 80% with one confidence interval in the abstract and results, and 80.7% with another in the table; the table value is the one that reproduces. The tobacco-dust figure of 94% does not reproduce from its own 22 of 24. The application audit is prospective in the abstract and retrospective in the Methods. And the recommended threshold differs between the abstract and the discussion.

Singh 2001. The Statistical analysis section calls the result "not highly significant" and the Discussion calls the reduction "definitely decreased". The headline resource figure divides three events that occurred in the 1997 control group by the 1998 study group's denominator.

Natkunarajah 2003. One Results sentence, "No tears were identified using the slitlamp alone", contradicts the sentence three lines above it reporting 16 of 18 identified at the slit lamp.