A survey on Artificial Intelligence and Big Data utilisation in Italian clinical laboratories

Section 1: General characteristics (questions 1–6)

A total of 227 (17%) of 1,351 SIBioC participants, working in a total of 484 clinical laboratories of Italy, responded to the survey, with 30% participation among the audited laboratory directors.

The biologist was the most prominent professional, followed by lab physicians and technologists (Figure 1).

Figure 1:

Professional profiles of respondents.

Within an age ranged from 20 to over 70 years, dividing the population into classes of 10 years, the majority of the interviewees were aged between 51 and 70 years (n=89, 39.2%; 63 females, 26 males). Female sex (n=143, 62.9%) prevails in all age groups (Fisher’s exact p=0.021), except for the 31–40 age group (43.7%).

Interviewees worked primarily in public hospitals’ National Health Service Laboratories (42.7%) or in hospital-universities (33.9%). Others worked in private laboratories (10.6%), in private hospitals or nursing homes (9.3%), in research hospitals (IRCCS, 1.3%), in diagnostics companies (0.9%) or in other non-clinical laboratories (1.3%).

Replies were received from all the 20 Italian regions as shown in Figure 2.

Figure 2:

Region of Italy where the laboratory/facility is situated.

Participation in the survey collected data from laboratory facilities of various sizes regarding the usual number of examinations performed annually. The participants were evenly distributed between laboratories performing less than 1 million examinations per year (29%), between 1 and 3 (26%) million and between 3 and 6 million (31%), while a small proportion were in facilities performing more than 6 million examinations per year (14%).

Section 2: Adequacy of digital equipment (questions: 7–15)

This section examined the adequacy of technological resources, a fundamental requirement for Big Data analysis. Less than half of the participants (41%) reported a ratio of 1:1 of personal computers to employees, while in more than 30% of the responses only 1 PC per 3 employees or less was reported. Corporate Wi-Fi is available in almost two thirds of cases (66%).

The availability of webcams and microphones with the possibility of installing the necessary software for online meetings was restricted in most cases to a small number of workstations (62%) and was even completely absent in 20% of the cases. There was disagreement over the adequacy of the connection in terms of speed and stability, and there was a general lack of satisfaction with respect both to the quality of software and to hardware equipment (Figure 3).

Figure 3:

Judgement of the quality of connections and adequacy of software and hardware.

In this first set of questions concerning the adequacy of digital equipment, no differences were found between the answers received from public or private institutions, and the answers did not depend on the number of tests performed by the laboratory either.

In the lab, just approximately a quarter (26%) of the participants regularly use IT solutions based on Cloud systems. Among the answers received, there are no differences between those from different professional categories (biologist, physicians and technologists) and between laboratory directors and the others.

Among those who have indicated the use of a cloud platform (188/227), Corporate Clouds (41%) and Google Drive (38%) are the most popular platforms, followed by Dropbox (9%) and Microsoft One Drive (8%) and others (4%).

Only for a percentage of 25% of the respondents, the Cloud is also used as an integrated work environment, utilising the online versions of the offered software and services.

Section 3: Access to health data (questions 16–22)

When the possibility of accessing patient health data was investigated, it is interesting to note that almost two thirds of the 208 respondents (60%, 126 answers) do not have access to any data other than laboratory data (Figure 4).

Figure 4:

Limitations on access to non-laboratory health data and what kind of data are retrievable in those cases where it is allowed.

There appears to be a general inability to access patients’ non-laboratory health data, which evidently does not depend on credential restrictions reserved for particular professions. In fact, this problem is also evident when filtering the sample for the Laboratory Director population (58% of them have no access).

Among those who have access to other non-laboratory health data, data related to the request for laboratory tests (diagnostic query/exemption) (52%), the hospital medical record (34%) or the electronic health record (14%) are mentioned as the types in order of frequency. Concerning the way the data are accessed, in half of the cases (53%) access is always possible with one’s own credentials, in the other half (46%) access is only allowed to authorised laboratory staff, and in a few cases data are provided only upon request to special facilities outside the laboratory (1%).

In addition, the degree of integration of laboratory data collected in the LIS with other patient health data collected in the company databases was examined. By “integration” is meant the connection between different data that the company obtains from its various internal sources. This integration has been defined as either total (encompassing laboratory diagnostic, imaging and clinical data) or partial (covering only LM branches, such as clinical chemistry, microbiology, genetics, haematology and immunohaematology). The majority of respondents (63%) indicated the existence of some sort of integration of health data; however, among the 209 responses received, it was mostly partial integration (105 responses, 80%) rather than total integration (49 responses, 20%). A considerable percentage of respondents are not aware of the level of integration in their company (30 replies, 14%).

Almost half of the interviewees (59 out of the 125 responses received to this question, 47%) were unable to offer any information on the type of platforms for storing integrated data, while from the responses received, data were integrated in the corporate data warehouse in the majority of cases (57 responses, 46%), and only in a few cases (9 responses, 7%) on cloud platforms.

In addition, the presence of total integrated data analysis systems in companies was investigated, such as software that, through web pages or applications installed on computers, allow simple or complex analyses of data from various sources within the company. Nearly half of the 197 answers in this field are ambiguous (“don’t know”, 95/197, 48%), followed by the disconfirmation of the existence of such software (“no”, 71/197, 36%), while only in a small percentage of cases is the presence of software for descriptive analysis only (17/197, 9%) or for both descriptive and predictive analysis (14/197, 7%) reported.

Narrowing the field to software for analysing laboratory data, there was an increase in affirmative responses with the presence of software for descriptive analysis in 30% of cases (58 out of 196 responses) and for both descriptive and predictive analysis in 16% of cases (31 responses). In 22% of the responses (44), the absence of any such software was reported, and still 32% of the respondents (62) were unaware of the situation in their company.

Section 4: Laboratory data management and analysis (questions 23–25)

LIS data management was examined in the fourth section. More than half of the users (63%, 123 out of 195 responses to this question) are able to extract data from the LIS on their own, mainly using preset methods (42%) and only in one fifth of situations using a free, configurable method (21%). Other times, data extraction requires contacting IT support (28%) or in extremely rare circumstances data extraction is not possible (2%), or respondents are not aware about data extraction techniques (7%). Those who are able to autonomously extract data (63%) were asked in the subsequent question to assess the suitability of data extraction, in terms of speed and volume of extracted data, for the purpose of “analysis with Artificial Intelligence methodologies”: extraction speed and volume of data do not satisfy 73 and 68% of the respondents respectively (score 1–2). The ratings given to the speed and volume of extraction from the LIS are shown in Figure 5.

Figure 5:

Judgements on speed and volume of data extraction from the Laboratory Information System.

With regard to the tools available for data analysis, several answers could be selected. The majority have an integrated functionality in the LIS (61%) and use spreadsheets (49%). Only 16% refer to the use of more specific software. The most frequently used programmes among those mentioned by the respondents are MedCalc, R, Analyze-it, SPSS, GraphPad and Phyton. As R and Phyton are the most suitable software for analysing Big Data (BD), it is important to underline that only a very small number of respondents actually use them (only two answers for Phyton and four for R).

Section 5: Big Data and Artificial Intelligence (questions 26–34)

Only 20% of 183 respondents stated that they had a good (15%) or outstanding (5%) understanding of the BAI subject matter, and an even smaller percentage (12%) felt they had a good (9%) or excellent (3%) level of competence in BAI methodologies.

Among the possible applications of Artificial Intelligence, respondents gave greater consideration to expert systems for the release of lab results and the reference intervals estimation, while less importance was given to predictive algorithms, image recognition, language processing. In fact, the latter two (language and image comprehension) are currently less felt than in the field of radiology, although there are applications in haematology and urinary LM [20, 21].

In the majority of laboratories (91%, 166 out of 182 replies) there are no ongoing projects involving the use of BAI. Moreover, in most laboratories (62%, 110 out of 178 answers) there is not a professional position with expertise in the field of statistical solutions applicable to Big Data. However, almost a third (28%, 49 out of 178) declared to have active external collaborations with computer scientists. Only in 10% of the cases (19 out of 178) are there laboratory staff with specific expertise on BAI.

Similarly, specific software, platforms and PCs dedicated to BAI analysis are almost absent in the laboratories (145 replies out of 180, 80%), in a percentage of cases external partnerships are engaged (29 replies, 16%), while only seven stated that dedicated software is available.

The questionnaire examined the management of studies in which it is not necessary to maintain the patient’s identity and on any anonymization procedures, with regard to the problem of privacy management in the use of large amounts of data. Most of the 170 respondents (60%) reported using a home-made anonymization procedure, while in 40% of cases informed consent is still required. No one reported the use of a standardised anonymization procedure.

The majority of the 170 respondents (91%) anticipates that AI will be used in LM to support professionals in their decision-making, whereas only a very small percentage (2%) envisages that the laboratory professional may eventually be replaced in some tasks, such as reading of peripheral blood smears or urine sediments. Only a tiny portion of the interviewees expects a marginal role (2%) or does not know what a conceivable role (4%) for AI in LM could be. Additionally, several interviewees anticipate that the AI might facilitate LM Research (free answer not provided among the options).

The most common responses, among the 168 received, to the question of what is the main barrier to the application of BAI, identified inadequate IT infrastructure and lack of specialised software as the major obstacles, followed by poor integration between different data sources, lack of expertise and difficulties in accessing data. Respondents, however, generally agreed that there is not one single obstacle to focus on, but rather that all the challenges mentioned are actually obstacles to the use of BAI in Italian clinical laboratories (Figure 6).

Figure 6:

Opinion on the major barriers to the implementation of Big Data and Artificial Intelligence.

Finally, the vast majority of participants (95%, 162 out of 170 respondents) expressed interest in education on LM BAI. The majority of respondents said they would like to improve their skills in data management (35%) or data analysis (37%), while a third said they would benefit from attending a course on the subject (28%).