Study characteristics
We included 13 studies that evaluated the use of telecytology for thyroid FNAB assessment, accounting for a total of 3856 thyroid cases [
2,
13,
14,
20,
23,
32‐
39]; four studies [
34‐
36] did not focus exclusively on thyroid nodules, but also assessed samples from other organs (such as lymph nodes, lung, liver, pancreas, head and neck, and salivary glands). Ten studies were retrospective [
2,
13,
23,
32‐
38], two were prospective [
14,
39], and one consisted of two different phases: the first being retrospective, the second prospective [
20]. Five studies were conducted in the USA [
32‐
34,
37,
39], three in Greece [
13,
23,
38], two in Turkey [
20,
36], one in Colombia [
35], one in Portugal [
2], and one in Brazil [
14]. All the studies were published between 2009 and 2022, with eight of them published after 2018 [
13,
14,
20,
34‐
37,
39].
Five studies used real-time microscopy [
14,
32‐
34,
39], four used static images [
13,
23,
36,
38], three used virtual slide technology/whole slide images (WSI) [
2,
20,
35], and one used both real-time and virtual slide technologies [
37]. In one study [
23] the slides were transmitted to a password-protected account where the cytopathologists could view them, one used a hospital website [
39], two of them used NetCame software [
32,
33], one used Caseviewer [
20], one used Skype [
14], one used WhatsApp [
36], one used a Cloud-based portal software [
35], three used a Web Browser [
13,
34,
38], one used an NDP server [
2], and one did not define the modality of transmission [
37].
Twelve studies examined the diagnostic concordance of TC diagnosis with final cyto-pathological diagnosis [
2,
13,
14,
20,
23,
32‐
38], seven analyzed TC application on ROSE for preliminary adequacy assessment of the samples [
14,
20,
32‐
34,
37,
39], four evaluated TC image quality [
13,
14,
35,
38], two examined the use of TC in preliminary diagnosis [
20,
32] and two evaluated the use of TC in acquiring a second opinion [
13,
35]. Four studies reported the diagnostic time when TC was used: Khurana et al. [
33] estimated 4 to 6 min for each slide pass using real-time microscopy; Yao et al. [
37] reported that the average time spent per slide was 270 s (about 4.5 min) using VisionTek Digital Microscope (VDM) real-time microscopy and 122 s (about 2 min) for single Z stack digital scan (SZDS) virtual microscopy, compared to 113 s (about 2 min) for conventional light cytology; Sahin et al. [
36] reported an average time of 4.25 min for image capture of thyroid FNAB and 2.14 min from image transfer to diagnosis using static images; Mosquera-Zamudio et al. [
35] reported that the average time spent to read and interpret a case (including relevant metadata of the patient) was 6.2 min using virtual microscopy.
A comprehensive view of the characteristics of the included studies is provided in Table
3S in the Online resource. Table
1 summarizes the main results of this review.
Table 1
Summary of results
Diagnostic concordance between diagnosis via telecytology and conventional cytology | The concordance rate between TC and conventional cytology was excellent in 7/11 (63.6%) studies. Most disagreements were found among the indeterminate and follicular neoplasm cases. Low-cost methods showed good concordance | Moderate |
Preliminary assessment of adequacy of samples | TC can be used to perform preliminary assessment and the non-diagnostic rate drops when adequacy of samples is evaluated with TC. The concordance rate of the adequacy assessment of samples between TC and conventional cytology was generally high, ranging from 74 to 100% | Moderate |
Evaluation of telecytology image quality | Image quality was referred to as perfect or nearly perfect in most cases, regardless of telecytology technique | High |
Diagnostic concordance between diagnosis via telecytology and conventional cytology
Twelve studies reported data on concordance rates [
2,
13,
14,
20,
23,
32‐
38]: four studies used static images [
13,
23,
36,
38], four used whole slide images (WSI) [
2,
20,
35,
37], and five used real-time TC [
14,
32‐
34,
37]. Nine of them compared TC with conventional cytology [
2,
14,
23,
32‐
37], one with histology [
20] and two had data on concordance with both conventional cytology and histology [
13,
38].
To evaluate the magnitude of the concordance rate, the values of kappa statistics were interpreted as follows: poor agreement (k between 0.00 and 0.20), fair agreement (k between 0.20 and 0.40), moderate agreement (k between 0.40 and 0.60), good agreement (k between 0.60 and 0.80), and excellent agreement (k between 0.80 and 1) [
35,
40]. Three studies reported results only with the percentage of agreement, without calculating the k statistic, but all values were above 90% of agreement and were considered excellent.
The concordance rate between TC and conventional cytology was excellent in 7/11 (63.6%) studies [
13,
23,
32‐
34,
36,
38], good in one (9.1%) study [
14], and moderate in one (9.1%) [
2], with only one (9.1%) study reporting a poor concordance rate [
35]. The study from Yao et al. [
37] reported two different intraobserver agreements from the two cytopathologists involved in the study (with the intraobserver agreement being the concordance rate between two different techniques performed by the same pathologist): the cytopathologist expert in TC had an excellent concordance rate with a k between 0.85 and 0.93 depending on the technique used (real-time TC and WSI respectively), whereas the other cytopathologist, with less exposure to digital pathology, had a good concordance rate of 0.7 and 0.75.
Some studies also analyzed the diagnostic categories independently, with the categories being benign, indeterminate, follicular neoplasm, suspicious for malignant, and malignant [
2,
13,
14,
32,
34]. Most disagreements were found among the indeterminate and follicular neoplasm cases, whereas the other three categories always resulted in a good or excellent concordance rate.
If we divide the studies according to the TC technique used, static image studies [
13,
23,
36,
38] always had an excellent concordance rate. Three (60%) real-time TC studies [
32‐
34] reported an excellent concordance rate and one (20%) [
14] found a good concordance rate; additionally, Yao et al. [
37] showed excellent or good agreement when the expert or the less skilled cytopathologist was involved, respectively. Among the studies that used WSI, the data varied: Yao et al. [
37] found an excellent or good concordance rate depending on exposure to TC (k = 0.93 for the expert and k = 0.75 for the other); Gerhard et al. [
2] found a moderate concordance rate, while Mosquera et al. [
35] found a poor agreement.
The interobserver agreement, meaning the concordance rate between different cytopathologists on the same TC diagnosis, was reported in 3 studies with different results. Georgoulakis et al. [
23] used static images and found an excellent interobserver agreement. Additionally, when the analysis was conducted in serial rounds, where the cytopathologists analyzed the same slides multiple times, the interobserver agreement, even if always excellent, improved slightly over time from a mean k = 0.89 in the first round to a mean k = 0.91 in the third one. Yao et al. [
37] and Gerhard et al. [
2] used the WSI technique and reported a good (k = 0.70) and a moderate interobserver agreement (k = 0.57) respectively; the same study from Yao et al. [
37] also used real-time TC and found a moderate interobserver agreement (k = 0.47).
Three studies [
13,
20,
38] compared TC diagnosis with the final histological diagnosis and always showed good reproducibility between the two techniques. More importantly, the accuracy rates with TC (sensitivity, specificity, positive predictive value, negative predictive value, and total accuracy) were comparable to those found with conventional cytology: Archondakis et al. [
38] measured accuracy of diagnosis based on static images and conventional cytology compared with post-thyroidectomy histological diagnosis, and found that the two accuracies were similar (with a total accuracy of 97.21% and 97.11% respectively); Canberk et al. [
20] compared their results with the one obtained in large series using conventional cytology (Bongiovanni et al. [
41] and Lee et al. [
42]) and found very similar results (total accuracy of 94% versus 69% and 95% respectively).
Additionally, the study from Canberk et al. [
20], which used TC for primary diagnoses of thyroid FNAB, was able to compare its results with the benchmarks of the Bethesda System and found a distribution of cases (benign, indeterminant, follicular neoplasm, suspicious malignant, and malignant) and a total surgical rate in line with the suggested annual benchmarks.
Two studies [
14,
36] evaluated low-cost methods of TC, meaning they only used a smartphone to capture the images and transmitted them with free software (one used “WhatsApp”[
36] and the other “Skype” [
14]). Sahin et al. [
36] used static images as TC technique, while Costa et al. [
14] used a real-time ROSE assessment. In both cases, they found a good concordance between TC diagnosis and conventional cytology, with a respective intraobserver agreement of 78.85% (k = 0.839) and 83.3% (k = 0.685).
Preliminary assessment of adequacy of samples
Seven studies reported data on the use of TC for preliminary assessment of adequacy of thyroid FNAB samples [
14,
20,
32‐
34,
37,
39]. They generally showed that TC can be used to perform preliminary assessment and that the non-diagnostic rate drops when adequacy of samples is evaluated with TC. Five studies [
14,
20,
33,
34,
37] showed that the concordance rate of the adequacy assessment of samples between TC and conventional cytology was generally high, ranging from 74 to 100%.
Using WSI, TC Canberk et al. [
20] reported an interobserver agreement rate between TC (performed by a cytotechnologist) and conventional cytology (performed by a pathologist) of 88% (24 out of 25 cases). When real-time microscopy was used, Khurana et al. [
33] reported that the number of cases that were considered unsatisfactory by TC was 21 out of 100: 4 out of the 21 nodules were considered benign on final cytological assessment and the remaining 17 cases were categorized as unsatisfactory also on conventional cytology. Trabzonlu et al. [
34] made a two-phase study. During the “test” phase, cytopathologists were made comfortable with the process and could view the slides in the same room by a TV screen without communicating with one another, performing only the adequacy assessment. During the second phase, diagnostic categories and specific diagnoses were required. They reported a concordance rate of 83.3% in the first phase of the study and 94.8% in the second one.
Lin et al. [
39] showed that the unsatisfactory rate dropped significantly (from 8.8% to 3.8%) when telecytology ROSE was introduced in a center that didn't perform preliminary assessment of FNAB. Additionally, there was no difference in the Unsatisfactory/Non-Diagnostic rates between TC ROSE and conventional ROSE. Izquierdo et al. [
32] also reported a drop in the unsatisfactory rate in the group in which samples were transmitted with TC for adequacy assessment compared to those which weren’t, but it didn’t reach statistical significance.
Yao et al. [
37] evaluated how the intraobserver concordance rate varied with real-time VDM, or virtual slide SZDS compared to conventional cytology. The comparison was made between two different cytopathologists; the intraobserver agreement rate was 0.94 and 0.74 between conventional cytology and VDM, whereas it was 1 and 0.86 between conventional cytology and SZDS respectively for cytopathologists A and B.
Evaluation of telecytology image quality
Four studies analyzed image quality, using static images [
13,
38], WSI [
35], and real-time images [
14]. In all cases, the quality was subjectively evaluated by the cytopathologists who performed the diagnosis. Image quality was referred to as perfect or nearly perfect in most cases, regardless of telecytology technique. The three studies with digital images evaluated image quality on a scale of 1–10; in the two studies with static images, all the participants reported a really high image quality, with a mean score of 9.5 [
13] and 9.1 [
38] respectively, while Mosquera et al. [
35] reported an average image quality score of 8.3 for WSI without Z-stack and 8.7 if Z-stack was used.
Costa et al. [
14] described the image quality of low-cost real-time telecytology. They analyzed the delay in image transmission, clarity of the image, and clarity of voice command, classifying the image as “excellent” when all three parameters worked flawlessly, “good” when only one parameter presented issues, and “poor” when there were problems with two or more parameters. Given that, the study reported an excellent quality of transmission in 57% of cases and a good quality in 24% of cases, with only 19% of cases reported as poor. Additionally, they examined differences in the concordance rates depending on the quality of transmission: poor, good, and excellent quality of transmission respectively had a concordance rate of 62.5% (k = 0.500), 77.8% (k = 0.625), and 88% (k = 0.774).