The impact of interventions applied in primary care to optimize the use of laboratory tests: a systematic review

Search strategy

The search was implemented in two electronic databases: Medline (1st January 2010–20th September 2019) and Embase (1st January 2010–20th September 2019). The collection period was limited from 2010-to the present due to the two “2015 reviews” [3], [4] considered to cover the previous period (1947–February 2014) and because this review is meant to give an overview of the interventions applied recently, since the use of the IT-requesting systems has supplanted the use of paper request forms used in the past. Nevertheless, studies reported in the two 2015 reviews which fulfill our inclusion criteria were also included and evaluated in the present review.

No limits on language, subject or type were placed on the database searches. The searches were designed to be comprehensive and to cover a broad range of the field, including the hospital setting, to ensure that all types of connections or collaborations with primary care appeared in the search. The search query consisted of terms considered important by the authors to describe the subject of the review. The search was tailored to the specific requirements of each database (Supplemental Material 1).

All citations were imported into the Endnote X9 citation manager with duplicate citations being removed automatically or manually when found later in the process. Citations were then imported into the web-based review software program Covidence [5] for subsequent title and abstract relevance screening and data characterization of the full articles.

All potential papers underwent a two-step screening process: (1) all citations were screened based on the title and abstract by two independent reviewers (SL, SA); (2) the full text of each citation was read and assessed for inclusion by the same two reviewers independently. Disagreements were solved by discussion.

Criteria

Studies were included in the review based on the following inclusion and exclusion criteria:

Data collection

For each study reported in this review, we extracted information regarding the authors, country, year of publication, evaluation period, type of interventional strategy(ies) used, laboratory test(s) evaluated, study design and outcomes in terms of:

• number of tests requested before applying intervention(s);

• number of tests requested after applying intervention(s)

• difference in number of tests before and after intervention(s)

We evaluated the quality of the studies as the overall risk of bias translated to “high risk of bias”, “medium risk of bias” or “low risk of bias”, based on the Cochrane Effective Practice and Organization of Care (EPOC) suggested risk of bias [6] (Supplemental Material 2).

The study designs included: Randomized control trials (RCT) and cluster randomized controlled trials (CRCT); non-randomized control trials such as before and after controlled studies (CBA); before and after studies and prospective studies lacking a control group or control test (BA); and interrupted time series analysis (ITS) [7].

All the information was stored and handled in an Access database 2007–2010.

Calculations

The effectiveness of the interventions was reported as a relative number, corresponding to the effect on the group of tests following the application of the intervention.

Studies describing more than one intervention, were reported in this review with two or more results.

When the difference in tests used was reported as percentages, the study result was used directly (for example 20% being identical to 0.20).

Otherwise, if the results of a comparator (control group or test) was also reported in the study (before and after the intervention was applied), we adjusted the change in the intervention group with the change in the comparator before further calculations:

1. Calculating adjustment for intervention group before, based on changing in comparator:

2. Calculating relative change in intervention group adjusted by the change in the comparator:

(Intervention group after− Adjusted interventiongroup before)/Adjusted intervention group before

In case of the intervention applied on more tests in the same study, a weighted relative change was calculated based on the absolute number of the individual tests both for the comparator (before and after) and the intervention group (before and after). The weight of each tests was simply the number of the individual test:

Weighted result = Σ [changes (adjusted) in intervention group * number of test intervention group (after)]/total number of tests in intervention group (after)

Types of intervention

The interventions used in the included studies were categorized as following:

Study selection

The electronic searches produced 9,985 studies (4,126 in Medline and 5,859 in Embase). After removing duplicates, 9,287 were screened at the title and abstract level. Four hundred and twenty-five articles were selected for full-text screening and of those, 332 studies were eliminated as they were not reporting original quantitative results relevant for the review topic. Eighty studies were excluded because the interventions were not applied to primary care.

The electronic search ended with 13 studies that met our pre-specified criteria and were included in the review. Seventeen studies were extracted from the two 2015 reviews [3], [4] and of those, six studies were excluded: three due to a different focus area, two because they were also included in our own search and one because no clear results were reported.

The selection process left a final pool of 24 studies (Figure 1) that were published between 1992 and September 2019.

Figure 1:

PRISMA Flow chart.

Studies selection.

Study characteristics

Table 1 summarizes the characteristics of the included studies. Five studies implemented more than one strategy to reduce the number of tests either to different pools of tests [8], [17], [26] or to the same pool of tests [12], [19] and therefore, the results of those studies have been split into two or three parts, respectively. Two studies used the same interventions in two different laboratory areas, and the results were therefore also reported in two sections [11], [28].

Table 2 includes studies that reported interventions applied for increasing the use of tests. One study investigated exclusively interventions aiming at increasing the use of (under-requested) tests [31] while three studies [8], [13], [24] reported, as a part of their investigation, both results related to over-use of tests as well as to under-used tests and therefore those results have been described separately in Table 2.

The studies included in this review were conducted in UK [12], [14], [22], Italy [27], [28], [30], The Netherlands [9], [10], [11], [20], Canada [26], USA [16], Israel [21], [23], [24], Sweden [13], Portugal [8], Spain [17], [18], [19], [29], [31], Finland [25], and New Zealand [15]. Although most were published in English, one was in Italian and another in Dutch.

The observational period in the 24 studies varied from three months to five years.

Study design and risk of bias

For 23 of the studies, the risk of bias was determined by following the criteria for RCT and non-RCT(6). Seven studies had a high overall risk of bias, of which five were BA [21], [22], [24], [26], [31], and three CBA [15], [23], [27]. Thirteen were classified as having a medium risk of bias, of which 11 were CBA [9], [13], [17], [18], [19], [20], [25], [28], [29], [30], [31], and two were RCT [10], [14]. Only three studies were classified as having a low risk of bias, one was a CRCT [12] and two were RCT [8], [11].

To determine the quality of the Interrupted time series (ITS) study, the ITS studies criteria was followed [6], and the study [16] was evaluated as having a medium risk of bias.

Of the five RCT/CRCT studies, randomization was performed at the clinics level in three of the studies [10], [12], [14], at the server level in one [8] (which means that all physicians using the same server received the same intervention) and at the primary care physicians’ level in the last one [11].

Of the 12 non-RCT studies, five used a test as control [17], [18], [23], [25], [29], and seven used a control group [9], [13], [15], [19], [20], [27], [28]. Five studies used a time period as comparator [21], [22], [24], [26], [30], while one prospective study did not have neither control group/test nor control period [31]. The ITS study [16] used a control group as comparator.

Intervention components

The majority of the studies (66%) used a single intervention approach, while the remaining studies used a combination of interventions.

Results obtained by different types of strategies combined with the risk of bias are displayed graphically in Figures 2 and 3. It is clearly illustrated that the risk of bias is lower for studies investigating the educational strategies, showing relative changes of approximately 10–30% reduction in test numbers. The medium risk of bias studies were represented in all categories and showed results ranging from approximately +15% to −90%. The high risk of bias studies were mainly included in the administrative category and the range of results were very broad going from +94% to −72%.

Figure 2:

Thirty-one results extracted from 23 studies that report the effects of interventions with the purpose to reduce the number of tests requested.

Some of the results were extracted from the same reference.

Figure 3:

Four results extracted from four studies that report the effects of interventions with the purpose to increase the number of tests requested.

Some of the results were extracted from references that also reported interventions with the aim to decrease the number of specific tests.

Studies published before 2010 reported only educational strategies (half of them supplemented by feedback reports) or administrative changes, while studies published later than 2010 implemented more technically complicated interventions (Table 3).

Impact of interventions aiming to decrease the number of tests

Impact of interventions aiming to increase the number of tests

Three studies were designed to apply interventions to increase the use of specific laboratory tests in addition to the strategies to decrease the use of other tests [8], [13], [24]. Only one study was designed to apply exclusively an intervention to increase the use of the evaluated test [31].

Results of interventions aiming to increase the number of tests are reported in Figure 3.

Studies designed together with requestors

Eight studies designed the intervention in consensus with the requestors [9], [16], [17], [18], [21], [27], [28], [31] and they are represented in all the intervention categories. The span of the results were very broad, ranging from +0.07 to −0.51. Moreover, none of the studies had a follow-up period.

Sustainability/applicability

High quality studies were primarily represented in papers evaluating educational components, and they often resulted in changes in test usage of 10–30%. The duration periods were however, generally short with most of the studies reporting interventions of less than one year, and most of the reports lacked a follow-up period, so it is unknown whether the changes are stable or the requesting patterns return to previous levels after the intervention period, with one outstanding exception: The long-term effect of the study of Larsson et al. [13] was evaluated after eight years [32] and showed that short continuation courses on optimal use of clinical chemistry can achieve permanent changes in the test ordering patterns of GPs.

One of the key roles of the laboratory is to provide an easy and direct access to guidelines and educational material for examining and treating the patients in the best possible way [33]. However, educational approaches often require a large amount of effort over time as well as dedicated staff and still may not be sufficient on their own [1], [34], [35].

We found only one study which applied a strategy where the cost of the tests are displayed in real time when the GP is ordering the tests [16], and it achieved insignificant results. In the hospital setting, this strategy has resulted in conflicting results, with some studies reporting that displaying the cost of the tests led to a decrease in the ordering of tests [36], [37] while other authors concluded the contrary [38]. Displaying costs on the GPs requisition form is still sparsely investigated, but GPs tend to use more common and cheaper tests compared to hospitals, which tend to request the majority of special and expensive tests. It is therefore likely that this intervention might have only minor effect on the requesting patterns of GPs.

Feedback reports on their own seem to have modest or no effect on requests made in primary care. This is in contrast to findings in the hospital setting, where the use of feedback reports (described in combination with educational components) have been found to have clear and prolonged effects on physician laboratory test-orders [39], [40]. However, as described by Pantojaet al. [41] the effect of feedback reports probably depend on a number of specific factors such as: How the reports are delivered to the GPs; whether they provide generic or specific information; whether they provide an explanation of their content or not; whether they require the healthcare professional to record a response or whether the targeted clinicians are involved in their development.

Many of the administrative interventions investigated alone or in combination with educational components were technically provided by the CPOE systems as changes of requesting profiles, new order profiles or unbundling panels.

Although administrative interventions achieved the greatest relative reductions, the range of the results differed widely, probably due to the heterogeneity of the studies and the high risk of bias of many of the studies, which makes the results uncertain. It was reported by the authors of the studies included in this review, that this kind of interventions needed relatively little effort to carry out and were easily maintained over time [17], [26] because changes in the IT-systems are permanent as well as independent of conditions in the individual clinics. However, as it is system-dependent, it may not be possible to apply everywhere.

The one study that used pop up alerts achieved a high reduction in number of tests requested [28]. In hospitals, this approach has been extensively used, both by applying non-interruptive alerts, which passively displays information, such as redundant testing or suggestions of corollary orders [28], [42], or interruptive alerts which require the user to acknowledge the alert [43], [44]. A theoretical benefit of these alerts is that they can provide “just in time” advice and thereby provide long-term education. Conversely, this type of interventions contain the risk of “alert fatigue” when the alert is displayed for long time periods, making the requester “blind” to the content of the alert and/or even irritated by the interruptions caused by the alert. This is probably most often seen, when alerts are used with tests that are requested in high numbers.

The use of reflexive tests or gate keeping rules have just recently been investigated among GPs (Table 3) both in regards to over-used and under-used tests and this intervention seem to achieve positive effects. The use of gate keeping whether it was combined or not with educational components obtained comparable results in the included studies, but we did not find a clear effect when reflexive testing was combined with education. Nevertheless, these advanced tools are described by the authors as interventions which are more suited when implementing clinical guidelines [19]. This is because laboratory professionals are in the position to manage demand on laboratory tests by utilizing evidence based guidelines, when developing specific test ordering directives and gate keeping rules [26].

According to the review of Cadamuro et al. [33], laboratory diagnostic algorithms are the most sophisticated tools to influence the appropriate use of diagnostic testing as they permit physicians to obtain relevant laboratory results without ordering individual tests and still arrive at a definitive diagnosis. It has been stated that this approach together with a comprehensive, patient-specific interpretation of the test results, can substantially reduce the cost by decreasing the time to diagnosis and the number of tests ordered without compromising patient care [39], [40], [41]. CPOE approaches are still only sparsely investigated in the primary care setting, however the studies included in this review are mainly medium risk of bias, suggesting that additional low risk of bias investigations are needed in the future.

In a number of studies, interventions were designed in collaboration and applied in consensus with the requestors [9], [10], [12], [16], [17], [18], [21], [27], [28], [31]. In some of this [16], [20], [28] the opinion of the GPs as to the applied interventions was investigated by questionnaires or interviews. However, none of the studies reported in this review have evaluated how the GPs view the impact of the different interventions on their daily practice. GPs are exposed to the intervention every day and, in the perspective of improving the laboratory test requests, having their opinion is essential. If the strategy applied is considered annoying and/or delaying the workflow by the GPs, the applied strategy should be re-evaluated. The tendency of applying strategies in close collaboration with the requestors seems to be an effective method to share information as well as potential problems that otherwise may have been overlooked, but based on the present results, we cannot substantiate this. Nevertheless, we think that a mutual discussion may bring better results or at least better implementation, and further studies in this direction is recommended.

Included in the limitations reported, there were short follow-up periods and poor study designs. However, the most commonly mentioned limitation was regarding the effect of the interventions on the patients. In all intervention studies, it was stated that it was unknown whether the reduced number of tests after intervention might have led to a missed diagnosis in patients. Evaluating appropriateness of test requests both before and after applying an intervention is a difficult task, especially when the clinical records of the patients are not integrated in the laboratory system. But, if the intervention is well designed and the appropriate use of the test under investigation is well documented and subject to clear clinical guidelines, changes in number of tests requested ought not to affect patient care.

Costs

The studies which reported changes in costs either as marginal costs or costs of reagents did not evaluate the cost-benefit of the interventions applied. Reporting only the reduction in costs without including any additional costs to the laboratory that might be necessary in regards to setting up and maintaining the intervention or informing the requestors, does not give a clear picture of the total expense involved. What could be done was to summarize the effect of how many blood test taken would potentially be taken if the intervention was not applied and consequently evaluate whether it was worthwhile to apply that intervention.

Limitations

Our review have some limitations. First of all, our search was only applied in two databases so potentially important studies included in other databases could be missed. In addition, the search terms did not include ‘demand management’ and it cannot be ruled out that some studies may have been missed on that account. Furthermore, because of the heterogeneity of the calculation of results in the different studies, we were not able to make a traditional meta-analysis, so our study is mainly descriptive. However, by adjusting the results, we were able to make the included studies comparable as to the relative changes in test numbers.