To analyze the service volume, turnaround time, and economic benefits of the CNCTPS, we collected all case data submitted from January 2016 to December 2019, which included demographic and clinical data, submitted hospital, case submission time, report issuance time, telepathology diagnosis, and specialist who made the diagnosis. After removing test cases, there were 23,167 cases. Specimens had been taken from multiple organs, which were divided into 26 groups.

To analyze the diagnostic accuracy of the CNCTPS, we followed up the final diagnosis of all the 23,167 cases through the hospital information system of the First Affiliated Hospital of Zhengzhou University. We searched and found that 564 cases had also been diagnosed directly in the First Affiliated Hospital of Zhengzhou University. The diagnostic accuracy of telepathology was calculated by using the final diagnosis in the First Affiliated Hospital of Zhengzhou University as the reference.

Descriptive statistics were used to analyze characterize case data, including demographic characteristics of patients from whom samples were taken, diagnosis, histopathology type, and turnaround time. The median value and interquartile range are reported for continuous data, and percentages are reported for categorical data. The Kruskal–Wallis H test was used to compare turnaround time in different years, and the Nemenyi test was used for further multiple comparisons. The concordance between CNCTPS and final diagnoses was analyzed (complete concordance or variance with no clinical significance). The consistency was determined by the McNemar test and consistency check. All statistical analyses were performed using R software (version 4.0.0; R Foundation for Statistical Computing). All tests were 2-tailed, and P<.05 is considered statistically significant.

During the 4-year study period from 2016 to 2019, 23,167 cases were submitted to the CNCTPS for consultation. The service volume of the CNCTPS was n=2335 in 2016; n=4330 in 2017; n=7262 in 2018; and n=9240 in 2019, with an average annual growth rate of 41.04%. A total of 83 hospitals participated in the telepathology consultation service. The number of participating hospitals also grew, from n=60 in 2016 to n=74 in 2019 (Figure 3).

Hospitals of different levels have joined the CNCTPS, including 17 city-level and 66 county-level hospitals. Among 2016 and 2019, city-level hospitals and county-level hospitals applied for 2880 (2880/23,167, 12.43%) and 20,287 (20,287/23,167, 87.57%) consultations, respectively. The number of county-level hospitals applying for consultation increased from n=49 in 2016 to n=63 in 2019, and the service volume also increased from n=2095 in 2016 to n=8317 in 2019. In city-level hospitals, the number of hospitals applying for consultations did not change, while the service volume showed an overall increasing trend (Table 1).

The locations, from which specimens had been taken, were divided into 26 groups (Table 2).

Of the 23,167 patients represented by case data, 9519 (41.09%) were male and 13,648 (58.91%) were female (Table 3). The median age of the patients, from whom specimens were taken, was 53 years (mean 52.86 years, range 1 day to 98 years). There were 17,495 out of 23,167 cases (75.52%) with confirmed diagnoses; 4779 out of 23,167 cases (20.63%) needed further examination, and most (4007/4779, 83.85%) required immunohistochemical examination. The other 893 (893/23,167, 3.85%) cases failed to be diagnosed, and poor slice quality and incomplete sampling were the main reasons thereof.

Among 17,495 confirmed cases, 12,088 were benign lesions, 5217 were malignant lesions, and 190 were borderline lesions. In total, 52.18% (12,088/23,167) benign cases and 22.52% (5217/23,167) malignant cases had been confirmed. The proportion of malignant lesions in the esophagus, lung/mediastinum, urinary, and thoracic cavity/pleura was higher than that of benign lesions (Figure 4). In the other 22 tissue types, the proportion of benign lesions was higher than that of malignant lesions.

The turnaround time, the time from transmitting whole-slide images to the issuance of diagnostic reports, was a median of 16.93 hours (IQR 32.59; mean 24.93 hours, range 100 seconds to 167.97 hours). Experts’ opinion reports were released within 12 hours in 10,244 of the 23,167 cases (10,244/23,167, 44.05%) and within 72 hours in 21,286 cases (21,286/23,167, 91.88%) (Table 4).

The difference in distribution of turnaround time in different years (Figure 5), was statistically significant (H=1433.62, P<.001). The median turnaround time gradually decreased in turn, from 29.36 hours in 2016 to 9.75 hours in 2019, and differences between subsequent years were statistically significant with pairwise comparison (between 2018 and 2019 adjusted P=.01, other groups adjusted P<.001).

Of 564 diagnosed by both the CNCTPS and pathologists in the hospital, 553 cases diagnosed by the CNCTPS were consistent with the final diagnosis made by pathologists in hospital; that is, the accuracy rate was 98.05%. In the other 11 cases—4 false-positive cases and 7 false-negative cases—5 of the 11 cases occurred in the uterus (Table 5).

The sensitivity and specificity were 97.66% and 98.49%, respectively (Table 6). The Youden index was 0.96. The positive and negative predictive values were 98.65% and 97.39%, respectively. No statistical difference was observed between telepathology diagnosis and final diagnosis (P=.55), which showed good consistency (κ=0.96, P<.001).

Telepathology consultation is free and avoids the need for patients having to visit higher-level hospitals. Therefore, consultation and travel costs were saved. At the same time, food costs were lower in the local area. Thus, compared with the traditional pathology consultation, diagnosis via the CNCTPS results in cost-savings of 378.5 RMB (approximately US $50) per patient (Table 7). In terms of the annual telepathology consultation cases, the total amount is substantial—approximately $300,000 per year.

The cloud-based system can quickly process data with large memory requirements, thereby overcoming the difficulties of large whole-slide imaging file management. This study reported on one of the largest cloud-based telepathology systems in China and evaluated its operation results. This study used a large sample size, which provides an in-depth practical understanding of the cloud-based telepathology system in China, and gives suggestions for further evaluations and improvements of the telepathology system. This system served 23,167 cases from 2016 to 2019. The median turnaround time was 16.93 hours, which decreased from 29.36 hours in 2016 to 9.75 hours in 2019. The diagnostic accuracy was 98.05%, and approximately $300,000 were collectively saved by patients each year. The CNCTPS has proven to be highly reliable and plays an important role in facilitating the distribution of limited senior pathologist resources in China.

A total of 83 hospitals are covered by the CNCTPS, which is the largest telepathology network in China. Compared with the 6, 24, and 60 workstations in other reported telepathology networks [23,24,26], the CNCTPS covers more medical institutions. Case data for more than 20,000 patients were diagnosed by the CNCTPS in 4 years. To the best of our knowledge, this is the largest sample size in a study on telepathology system use and operation and is much higher than those in similar literature [24,26-28]. The amount of case data reviewed and the number of participating hospitals increased each year, consistent with findings reported by Chen et al [26] and Zhou et al [6]. Most case data (20,287/23,167, 87.57%) had been submitted by county-level hospitals because the shortage of pathologists in China’s county-level hospitals is more severe than that in city-level hospitals.

A total of 893 cases failed to be diagnosed by the CNCTPS, of which only 26 cases were complicated enough that needed to be consulted in a higher-level hospital, while others were due to incomplete sampling and poor slice quality. Standard materials and good slice preparation are the main factors affecting telepathology diagnosis [29], which require experienced pathology technicians. Although we had conducted theory and practical operation training for pathology technicians in the early stage of CNCTPS construction, incomplete sampling and poor slice quality were still the main reasons for failed diagnoses. Strengthening the training of telepathology staff in the later stages of system operation is still needed. In terms of histopathology type, an analysis of ten-year telepathology cases in Tanzania showed a higher proportion of benign cases, which reported 56.1% benign and 40.8% malignant diseases [27]. We reached a similar conclusion: the proportion of benign cases (12,088/23,167, 52.18%) was higher than that of malignant cases (5217/23,167, 22.52%).

The average turnaround time of the CNCTPS was 24.93 hours, which is shorter than the 38 hours reported by Zhou et al [6] and 66 hours reported by Völker et al [27] but is slightly higher than the 0.7 days (ie, 16.8 hours) reported by Chong et al [24]. The majority (14,835/23,167, 64.04%) of cases were diagnosed within 24 hours, which is higher than the 61.5% reported by Chen et al [26] and slightly lower than the 64% reported by Perron et al [23], but the proportion within 48 hours (82.88% vs 70.00%) and 72 hours (91.88% vs 80.00%) were higher than those reported by Perron et al [23]. Nonetheless, the median turnaround time decreased annually during the 4 years, indicating that the CNCTPS operates well.

Compared to static images in the early stage of telepathology, whole-slide imaging allows the entire slide to be viewed in a manner that simulates microscopy [13]. A recent meta-analysis [30] shows that the weighted mean of the concordance rates between telepathology and conventional microscopy was 91.1% up to 2000, and from 2000 onward, the weighted mean of the concordance rates was 97.2%. It has been asserted that the reasons for increased consistency rate in recent years should be attributed to the increased use of whole-slide imaging [30]. The range of diagnostic concordance rates between whole-slide imaging and traditional electron microscopy is 89% to 100% [31-35], and the average value is 96.9%. Our study demonstrated similar results (98.05%). Moreover, no statistically significant differences were found (P=.55) between whole-slide imaging and traditional pathology diagnosis, and the consistency of diagnostic results was excellent, which further confirmed the accuracy of whole-slide imaging.

Some cost-effectiveness studies have demonstrated that telemedicine can reduce costs [36], but not all [37,38]. Cost-utility and cost-effectiveness studies for telepathology are rare. Meléndez-Álvarez et al [17] only evaluated the cost of their telepathology system, which saved US $410. Vosoughi et al [39] evaluated the cost-efficiency of their real-time nonrobotic telepathology system, which saved US $10,767.10 per year. In our study, cost savings for patients were estimated. During the 4 years of the telepathology system operation, approximately US $300,000 per year was saved by patients.

To the best of our knowledge, this is the first study to comprehensively evaluate the operation of a telepathology system based on a large sample. The CNCTPS showed fast responsiveness and high accuracy. However, owing to the limited information collected by the CNCTPS, this study did not analyze the reasons for cases with long turnaround time or the reasons for false positives and false negatives. In addition, only the costs saved for patients were evaluated in the economic benefits of the CNCTPS. The economic impact of telemedicine is a collaborative and complex process in which different economic, social, and political actors can be involved [38], and the construction of our system is a public welfare project initiated by the government and a powerful hospital. Most of the digital slide scanners were donated to the participating hospitals, and the private network was free.

Turnaround time and diagnostic accuracy are the main criteria used to evaluate a telepathology system, and further work is required to explore the factors that influence turnaround time and diagnostic accuracy. First, it is necessary to analyze the causes of cases with long turnaround time through a survey of pathologists, especially those with turnaround times longer than 72 hours. Second, more investigation for incorrectly diagnosed images is needed. In addition, adding a follow-up module to the CNCTPS is necessary to allow the final diagnosis result of each case to be easily followed up. Finally, a user satisfaction survey should be conducted, with thorough questionnaires or in-depth interviews, in a subsequent study to improve the system.

The CNCTPS has proven to be highly reliable. It can provide rapid telepathology diagnoses to participating hospitals that are consistent with the final diagnosis. The application of this system reduces financial costs and time for patients, facilitating the distribution of limited senior pathologist resources in China. Therefore, we believe telepathology services will become more widespread, in more regions worldwide, especially those with insufficient medical resources.