A total of 31 full text studies were identified for inclusion in the review. The search of Ovid MEDLINE, PsycINFO and PubMed databases provided a total of 675 citations. After adjusting for duplicates 469 remained. Of these, 392 studies were discarded because a review of the abstracts identified that these papers did not meet the criteria. The full text versions of the remaining 77 citations were examined in more detail. Of these, 46 studies did not meet the inclusion criteria as described. Thirty-one studies met the inclusion criteria and were included in the final systematic review.
Q1. Which adherence related metrics have been derived from prescription claims records?
Q2. What are the applications of adherence metrics for health outcomes?
All studies were of retrospective observational cohort design. The range of participants was 354 to 32374. Average number of participants enrolled in the included studies was 6491 with standard deviation (SD) ± 7512 across 31 studies.
The mean age across studies of the enrolled participants was 41 (SD= ± 7.4). The proportion of females varied from 32% to 69.2% with a mean of 52.8% (SD = ± 8.5%).
Clinical diagnosis of the enrolled participants included Schizophrenia and Bipolar Disorder (table 3 – see appendix 1). Most of the 31 studies were conducted on those diagnosed with disorders of Schizophrenia (21 studies), Bipolar disorder (6 studies), included both Schizophrenia and Bipolar Disorder (3 studies) and one paper reported on Severe Mental Illness (1 study).
All studies used prescription claims adherence measures to report the impact of non-adherence. Reported outcomes of non-adherence included psychiatric hospitalisation (27 studies), all cause hospitalisation (14 studies), cost & resource use (11 studies), mortality (1 study), and criminal arrest (1 study). The most reported outcome was psychiatric hospitalisation (table 1).
Table 1: Prescription claims-based adherence measures and outcome and associated health outcomes
|
|
|
Outcome measures
|
Adherence metric
|
|
|
|
|
|
N
|
Author
|
Year
|
PH
|
AC
|
CR
|
M
|
A
|
MPR
|
PDC
|
Gap
|
N
|
Sex - % Female
|
Age
|
SD
|
Funding
|
1
|
Becker, Young (16)
|
2007
|
|
|
1
|
|
|
1
|
|
|
10330
|
46.1
|
41.95
|
9.61
|
Not stated
|
2
|
Broder, Bates (17)
|
2012
|
1
|
|
|
|
|
1
|
|
1
|
2316
|
56.9
|
40.5
|
12.6
|
Bristol-Myers*
|
3
|
Broder, Greene (18)
|
2019
|
1
|
|
1
|
|
|
|
1
|
|
18388
|
69.2
|
40.3
|
14.5
|
Otsuka*
|
4
|
Cullen, McGinty (19)
|
2013
|
|
|
|
1
|
|
1
|
|
|
2132
|
46.5
|
41.6
|
10.1
|
Not stated
|
5
|
Eaddy, Grogg (20)
|
2005
|
1
|
|
1
|
|
|
1
|
|
|
7864
|
60
|
42.9
|
11.9
|
Janssen
|
6
|
Egglefield, Cogan (21)
|
2019
|
|
1
|
|
|
|
|
1
|
|
14,365
|
56.6
|
33.3% <45
|
NS
|
Not stated
|
7
|
Gianfrancesco, Sajatovic (22)
|
2008
|
1
|
|
1
|
|
|
1
|
|
|
8092
|
55.6
|
38
|
16.5
|
AstraZeneca
|
8
|
Gilmer, Dolder (23)
|
2004
|
1
|
1
|
1
|
|
|
1
|
|
|
1619
|
44.1
|
42
|
11
|
NIMH
|
9
|
Hassan and Lage (24)
|
2009
|
1
|
1
|
|
|
|
1
|
|
|
1973
|
63.9
|
39.5
|
13.2
|
AstraZeneca
|
10
|
Jiang and Ni (25)
|
2015
|
1
|
|
1
|
|
|
|
1
|
1
|
32374
|
60
|
51
|
15.3
|
Not stated
|
11
|
Joe and Lee (26)
|
2016
|
1
|
1
|
1
|
|
|
|
1
|
1
|
5548
|
NS
|
48.28
|
16.91
|
Not stated
|
12
|
Karve, Cleves (27)
|
2009a
|
1
|
1
|
|
|
|
1
|
1
|
|
3395
|
52.5
|
42.9
|
13.2
|
Not Funded
|
13
|
Karve, Cleves (28)
|
2009b
|
1
|
1
|
|
|
|
1
|
1
|
1
|
3395
|
52.5
|
42.9
|
13.2
|
Not Funded
|
14
|
Kozma and Weiden (29)
|
2009
|
1
|
|
|
|
|
1
|
|
1
|
1499
|
53
|
45.1
|
12.4
|
Janssen
|
15
|
Kuwabara, Saito (30)
|
2015
|
1
|
|
|
|
|
1
|
|
|
657
|
66.21
|
38.28
|
12.38
|
Janssen
|
16
|
Lage and Hassan (31)
|
2009
|
1
|
1
|
|
|
|
1
|
|
|
7769
|
64.17
|
39.91
|
12.74
|
AstraZeneca
|
17
|
Lang, Meyers (13)
|
2010
|
1
|
1
|
|
|
|
1
|
|
1
|
12032
|
52
|
43.2
|
13
|
Janssen
|
18
|
Law, Soumerai (32)
|
2008
|
1
|
1
|
|
|
|
|
|
1
|
1191
|
45.51
|
57.9% (35-54)
|
NS
|
Public + Private
|
19
|
Lee, Soh (33)
|
2018
|
1
|
1
|
|
|
|
|
|
|
4567
|
49.9
|
23.1
|
4.4
|
KMH*
|
20
|
Marcus, Zummo (34)
|
2015
|
1
|
1
|
|
|
|
|
1
|
|
3428
|
50.1
|
38
|
13
|
Alkermes
|
21
|
Markowitz, Karve (35)
|
2013
|
1
|
1
|
1
|
|
|
|
1
|
|
2541
|
43.33
|
41.17
|
12.18
|
Janssen
|
22
|
Offord, Lin (36)
|
2013a
|
1
|
|
1
|
|
|
1
|
|
1
|
1462
|
49.5
|
39
|
NS
|
Otsuka*
|
23
|
Offord, Lin (37)
|
2013b
|
1
|
1
|
1
|
|
|
1
|
|
|
354
|
65
|
71
|
NS
|
Otsuka*
|
24
|
Pilon, Tandon (38)
|
2017
|
1
|
|
|
|
|
|
1
|
|
24662
|
48.7
|
40.4 to 44.2
|
13
|
Janssen
|
25
|
Pilon, Alcusky (39)
|
2018
|
1
|
|
|
|
|
|
1
|
|
1410
|
32 to 37.1%
|
40.61
|
13
|
Janssen
|
26
|
Roberto, Brandt (40)
|
2017
|
1
|
1
|
1
|
|
|
|
1
|
|
13,681
|
NS
|
NS
|
NS
|
Not Funded
|
27
|
Svarstad, Shireman (41)
|
2001
|
1
|
|
|
|
|
1
|
|
1
|
619
|
NS
|
44
|
14
|
NIMH
|
28
|
Tang, Hsieh (42)
|
2010
|
1
|
|
|
|
|
1
|
|
|
4267
|
53.57
|
40.99
|
13.68
|
Janssen
|
29
|
Van Dorn, Desmarais (43)
|
2013
|
|
|
|
|
1
|
1
|
|
|
4056
|
55
|
37.3
|
12.4
|
FAHCA
|
30
|
Weiden, Kozma (44)
|
2004
|
1
|
|
|
|
|
1
|
|
1
|
4326
|
41.5
|
44.2
|
12.3
|
Janssen
|
31
|
Wu, Wang (45)
|
2015
|
1
|
|
|
|
|
1
|
|
|
920
|
47
|
39.3
|
13
|
FEMH*
|
|
Totals
|
|
27
|
14
|
11
|
1
|
20
|
11
|
10
|
|
|
|
|
|
|
Key PH = Psychiatric Hospitalisation, AC All Cause Hospitalisation, CR Cost & Resource Use, M Mortality, A Arrests, Bristol-Myers Squibb Company & Otsuka Pharmaceuticals*, Otsuka American Pharmaceuticals*, Korean Mental Health Technology R&D Project, Ministry of Health & Welfare*, Janssen Scientific Affairs, LLC*, NIMH -National Institute of Mental Health*, FEMH - Far Eastern Memorial Hospital, FAHCA - Florida Agency for Health Care Administration NS – Not Stated
Most articles (21 of 31 studies – 68%) reported a pharmaceutical funding sponsor. Janssen sponsoring 9 (29%) of the 31 full text studies (table 1).
Adherence measures used: MPR (21 studies), PDC (11 studies) and GAP (10 studies) were the most frequently reported adherence estimation techniques. MPR was the highest followed by PDC then GAP, however the GAP studies varied in the size of time-gap (e.g., 30 days, 60 days or even 90 days) and therefore was less consistently applied than the other two measures (table 2). How each measure is calculated is described elsewhere (15).
Table 2: Abbreviations and List of Adherence Measures
Abbreviation
|
Type of adherence measure
|
MPR
|
Medication Possession Ratio
|
PDC
|
Proportion of Days Covered
|
GAP
|
Gaps in Medication
|
NPMG
|
New Prescription Medication Gap
|
CR
|
Compliance Rate
|
CSA
|
Continuous Single Interval Measure of Medication Aquisition
|
CMG
|
Continuous Measure of Medication Gaps
|
CMA
|
Continuous Measure of Medication Acquisition
|
CMOS
|
Continuous Multiple Interval Measure of Oversupply
|
MRA
|
Medication Refill Adherence
|
RCR
|
Refill Compliance Rate
|
MPRm
|
Medication Possession Ratio, modified
|
DBR
|
Dates Between Fills Adherence Rate
|
Two of 31 studies reported the diverse outcomes of mortality (19) or criminal arrests (43) and were not comparable with any of the remaining 29 studies that mostly reported hospital or cost-based outcomes. These two studies were therefore excluded from subsequent analysis and are absent in table 3 (see appendix 1).
A widely used threshold for MPR / PDC non-adherence was <0.8 (17 studies - 55%). MPR is a calculation of how much medication a patient has in their possession (supply) divided by the days in a set period of treatment (days observed) (46). A patient would be stocked at 100% (1.0) if they possessed all the medication required for their health condition over a delineated period. If it falls below this figure (e.g., <80% or 0.8), then you can assume there are issues with the patient accessing their required supply. The threshold for defining non-adherence ranged from 0.5 – 0.8 for MPR, 0.4 – 0.8 for PDC and partial or non-fulfilment of a prescription was used for GAP. Ten studies detailed under the heading ‘adherence metric’ in table 1 (2, 10, 11, 13, 14, 17, 18, 22, 27, 30) used GAP as an adherence measure, seven of which included a GAP of 30 days. In half of these studies multiple GAP thresholds were used such as 1-10, 11-30, and greater than 30 days.
The time used in assessing non-adherence, detailed as ‘assessment time interval’ in table 3 (see appendix 1), varied between 3 months to 2 years: 11 studies used less than one year (2, 3, 7, 11, 15, 17, 18, 20, 27, 28, 31), while 17 studies used 1 year or more (1, 5, 6, 8-10, 12-14, 19, 21-26, 30) shown in table 3.
Studies used either a fixed approach to assess adherence from start to end of the time period for the whole cohort, or a person specific approach consisting of the start of the time period until an event occurred (e.g., hospitalisation) during the assessment period. Ten studies calculated adherence over a fixed time interval for all participants (studies 5, 7, 10, 12, 13, 21, 24, 26, 27, 30 in table 3 – appendix 1 ‘fixed or person specific’), while 19 studies used a participant specific time window to measure non-adherence as they measured adherence from index date until an outcome event occurred (studies 1-3, 6, 8, 9, 11, 14-20, 22, 23, 25, 28 and 31 in table 3). The time period used for observing outcomes of non-adherence varied from 6 months – 36 months: five studies used less than 12 months (2, 3, 7, 15, 20), 19 studies used 12 months (6, 8-14, 16, 17, 21-28, 30) and five studies (1, 5, 18, 19, 31) used longer than 12 months - table 3 – appendix 1 ‘follow up time period’.
The most common time used for assessing adherence in fixed approach studies and person specific approach studies was 12 months.
The time periods used for overall outcome observation varied from 6 months – 3 years: 24 studies used 12 months or more (1, 5, 6, 8-14, 16-19, 21-28, 30, 31), with 19 studies using 12 months (6, 8-14, 16, 17, 21-28, 30).
Adherence and outcome measures were calculated in 12 (41%) of 29 studies from consecutive time periods (non-overlapping) - table 3 (– appendix 1) provides outcomes observed in different time periods. Whether there was a temporal separation/gap between adherence and outcome measurement time was reported in 26 studies, 3 studies did not specify details (13, 26, 30). Of these in 14 (14/24 = 45%) studies adherence and outcome measures were calculated from different time periods (19, 20, 22, 25, 27, 28, 33-35, 40-43, 45). In 10 studies overlapping time periods were used for computing both adherence and outcome metrics (16, 18, 21, 23, 29, 32, 36-39).
The most commonly used model for reporting statistical relationships between non-adherence and outcomes was logistical regression table 3 ‘statistical model’, and effects were reported either as likelihood, odds or hazard ratio (1, 2, 5, 6, 7, 9, 12, 13, 14, 15, 16, 17, 18, 19, 31), rate of (3, 8, 11, 21, 27, 28, 30) or number of hospitalisations (20, 22, 23) table 3 ‘outcome’. An additional measure identified was the amount of health care resources utilised (1, 3, 5, 7, 8, 10, 21, 24, 25, 26, 27) table 3 (appendix 1) ‘outcome’.
Relationship between adherence and outcome
Of the 31 studies, 29 (94%) studies reported association between non-adherence and an adverse outcome - table 3 outcomes (appendix 1).
Twenty-four (83%) of 29 studies reported non-adherence were associated with the very common negative outcome of increased hospitalisation, and this did not change based on the adherence estimation methods used — (MPR, PDC or Gap), time window, fixed or person specific, with or without lag in the time periods between non-adherence and follow up. Across all reported follow up periods non-adherence was reported to increase hospitalisation, increase costs, or led to a detrimental outcome including criminal arrests and increased mortality. The link between non-adherence, hospitalisation and poor health outcomes is clear.