key: cord-0959852-x8b0ad7i authors: Edifor, E. E.; Brown, R.; Smith, P.; Kossik, R. title: Non-Adherence Tree Analysis (NATA) - an adherence improvement framework: a COVID-19 case study date: 2020-07-03 journal: nan DOI: 10.1101/2020.06.30.20135343 sha: 398316df21dcdb3913df0471e7e26801c5abf547 doc_id: 959852 cord_uid: x8b0ad7i Poor adherence to medication is a global phenomenon that has received a significant amount of research attention yet remains largely unsolved. Medication non-adherence can blur drug efficacy results in clinical trials, lead to substantial financial losses, increase the risk of relapse and hospitalisation, or lead to death. The most common methods measuring adherence are post-treatment measures; that is, adherence is usually measured after the treatment has begun. What the authors are proposing in this multidisciplinary study is a technique for analysing the factors that can cause non-adherence before or during medication treatment. Fault Tree Analysis (FTA), allows system analysts to determine how combinations of simple faults of a system can propagate to cause a total system failure. Monte Carlo simulation is a mathematical algorithm that depends heavily on repeated random sampling to predict the behaviour of a system. In this study, the authors propose the use of Non-Adherence Tree Analysis (NATA), based on the FTA and Monte Carlo simulation techniques, to improve adherence. Firstly, the non-adherence factors of a medication treatment lifecycle are translated into what is referred to as a Non-Adherence Tree (NAT). Secondly, the NAT is coded into a format that is translated into the GoldSim software for performing dynamic system modelling and analysis using Monte Carlo. Finally, the GoldSim model is simulated and analysed to predict the behaviour of the NAT. This study produces a framework for improving adherence by analysing social and non-social adherence barriers. The results reveal that the biggest factor that could contribute to non-adherence to a COVID-19 treatment is a therapy-related factor (the side effects of the medication). This is closely followed by a condition-related factor (asymptomatic nature of the disease) then patient-related factors (forgetfulness and other causes). With this information, clinicians can implement relevant measures and allocate resources appropriately to minimise non-adherence. A great proportion of patients (especially those with chronic diseases) are non-adherent to their 44 medication regimen [1, 2] . This has led many researchers to the conclusion that non-adherence poses a 45 significant challenge in medical practice [3, 4] . Some authors [5] class non-adherence as an 46 "epidemic", while the World Health Organisation (WHO) [1] considers it as "a worldwide problem 47 with striking magnitude". Patients' non-adherence to treatment interventions could have grave 48 consequences; it could blur the efficacy of treatments [6] , create large financial costs to sponsors [7] , 49 cause adverse events or even lead to death in some cases [8] . 50 Non-adherence to medications is not limited to any particular disease -acute or chronic; it affects all 51 diseases [9] and can be influenced by the timing, consistency and persistence of taking medications. 52 Barriers to medication adherence can vary significantly, ranging from patient-related barriers to 53 treatment-related barriers. Care providers, the healthcare system and medical staff also contribute to 54 non-adherence [4, 10] . Given this variation in barriers to adherence, there is no single intervention that 55 will effectively minimise medication non-adherence [4, 11] . For example, behavioural modification is 56 one way to improve adherence however, this is a very challenging solution to implement as human 57 behaviour is not easily altered. Behavioural modification can take the forms of education, motivation, 58 support and monitoring [12] . Tackling individual aspects of non-adherence can be done, however, 59 there is a need for a multidisciplinary approach to medication non-adherence [12] . 60 There are various techniques for assessing non-adherence. Though some are classic, such as pill 61 counting, others employ more sophisticated approaches [13] . Methods for measuring medication 62 adherence can be generally put in two main categories: direct and indirect [3] . The former provides 63 proof that patients have taken their medication as prescribed while the latter cannot provide such 64 proof. Direct methods include body fluid sampling, direct observation of patient and measurement of 65 biological markers [6] . Indirect methods, which are more widely implemented, include pill count, 66 patient questionnaire [14] , self-report forms, and electronic monitoring devices. Medication adherence 67 is characterised by three main components: initiation (the point when the patient takes the first dose as 68 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint prescribed), implementation (period of dosing regimen complying with prescription) and 69 discontinuation (the point when the patient stops taking medication as prescribed) [15] . 70 Measuring medication adherence can be challenging due to the use of adherence measures that have 71 poor accuracy and reliability [16] . Most of the methods for measuring adherence are performed during 72 the implementation phase of adherence [3] . Sometimes adherence measurements are performed 73 during the discontinuation phase [17] . There is limited literature on the methods for measuring 74 adherence before the initiation phase. Self-report methods of measuring adherence are usually 75 performed during the implementation phase. However, these self-reporting tools can be used as 76 historical data to measure adherence before the initiation phase of other future treatments. The 77 Medication Adherence Reasons Scale (MAR-Scale) and the Morisky Medication Adherence Scale 78 (MMAS) can be used to measure adherence before the initiation stage [18, 19] . Knowing the common 79 reasons for a patient's non-adherence to medications that they take for their chronic medical 80 conditions can help clinicians or pharmacists design interventions that will increase the chances of the 81 patient adhering to the new medication before the patient starting the medication. The MMAS scale 82 requires the patient to have other chronic medical conditions for which they are taking medications. 83 The MAR-Scale is unable to fully capture and analyse system conditions that may contribute to non-84 adherence but may not be directly associated with the patient or the medication. 85 Various techniques proposed for improving adherence are complex and ineffective, therefore, they are 86 unable to realise the full benefits a treatment could deliver [16] . It is rational to assess and measure 87 patients' likely non-adherence before the initiation stage of medication treatment to improve 88 adherence. The authors employ a proven probabilistic risk assessment technique to estimate the 89 likelihood of non-adherence before the initiation stage. The results of this study help clinicians to 90 identify and assess barriers to adherence; this aids them in the development of non-adherence 91 mitigating strategies and allocation of resources to improve adherence before the initiation stage of 92 medication adherence. 93 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. In Fig 1X, the output event Q is triggered when its input (or child) events A and B have occurred 106 within a given time, t. In Fig 1Y, the output event Q is triggered after a given time, t, when at least 107 one of its input events -A or B -have occurred. Fig 1Z depicts the PAND gate where the output 108 event Q is triggered after a given time t only when its input events A occurs before B. For a detailed 109 description of how FTA is performed, the reader is referred to Vesely et al. [20] . In a logical 110 expression, the AND, OR and PAND gates are represented by the symbols, *, + and < respectively. 111 Once a system has been translated into a fault tree, it can be analysed logically (qualitatively). The 112 logical analysis involves the determination of minimal cut sets (MCS) using Boolean algebra. MCS is 113 the smallest combination of basic events that are necessary and sufficient to cause the top event. 114 Necessary means each basic event in the MCS is needed for the top event to occur and sufficient 115 means the MCS does not need the occurrence of additional events to cause the top event occurrence. 116 In addition to creating MCS, the logical analysis also reveals single points of failure of a system and 117 reveals relationships between components. Quantitative analysis or probabilistic analysis involves the 118 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint To improve the overall reliability of the system, one could perform criticality/sensitivity analysis [20] 127 to determine how individual components contribute to the system failure. The results of a sensitivity 128 analysis enable investigators to implement mitigating strategies, know the quality of components to 129 use and allocate resources appropriately to improve the overall reliability of the system. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. This is the probability that a particular NAF will occur 190 and it is represented by P(NAF). The determination of the P(NAF) is based on the probability 191 distribution of the NAF. For example, given a duration (d) of 1 and 10 days respectively, if 192 FORG is exponentially distributed, P(FORG) can be evaluated as: 193 Non-Adherence Probability: This is the overall non-adherence probability -the probability 194 that there will be discontinuation as a result of NAFs at the end of a medication regimen. 195 Represented by P(NA), the non-adherence probability can be evaluated from Eqn 3. 196 Therefore, from the scenario, given a duration (d) of 10 days, P(NA) is 197 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. "there is less consideration given to adherence" [24] therefore, this study will only consider the out-206 patient settings. It is assumed that the treatment for out-patients, who usually have mild symptoms, is 207 a tablet that will be administered for 10 days by the patients themselves -one pill per day -for a study 208 population of 1000 patients. The diagram in Fig 3 is a NAT for a hypothetical treatment intervention 209 for COVID-19 using the WHO's dimensions [1, 5] and NAFs from six studies [25] [26] [27] [28] [29] [30] . 210 In general, there is a strong correlation between family/social support networks for patients and their 220 adherence to a medication regimen [31] . Patients with COVID-19 require self-isolation to avoid the 221 spread of the disease. Therefore, limited social support (SocSup) and limited healthcare access 222 (HeaAcc) have been considered as NAFs contributing to SocRel. Since this intervention is novel, it is 223 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint assumed that a NAF in the HeaRel category is "lack of prior knowledge of adherence" (PriKno) [1, 5] Non-Adherence = SocRel + PatRel + ConRel + HeaRel + TheRel 232 PriKno + SidEff + (IctSys * ManSys) 236 The MCS reveals that there are ten single points of failure in the system. Both IctSys and ManSys 237 need to occur together to trigger discontinuation therefore they are not considered single points of 238 failure. With a quick scan at these single points of failure, investigators can determine which aspect of 239 the system need backups. For example, NoTab is a factor that could be easily and quickly improved to 240 enhance adherence; not all the other factors can be quickly improved. For a detailed analysis on which 241 factor contributes most to non-adherence, probabilistic analysis is required. Probabilistic analysis can 242 only occur when NARs have been determined. 243 244 It is assumed that the recruited ambulatory participants would fail to adhere to their medication due to 245 HeaAcc resulting in a WNAR of 1.2E-4/day. It is also assumed that PriKno [1, 5] , has an initial 246 WNAR of 1.5E-4. This rate reduces by 8 per cent of the initial WNAR multiplied by the number of 247 elapsed day to represent the increasing knowledge of adherence by the medical team. The IctSys and 248 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint ManSys sub-systems responsible for ordering and dispensing the medicine fail at daily rates of 8.12E-249 5 and 5.34E-5 with mean-delay-time-until-repair of 4 hours and 2 hours respectively. There is a 2-day 250 delay time until the medication is delivered in case of NoTab. From results presented in Belmaker et 251 al. [25] , it is estimated that the WNAR (per day) for SocSup for patients taking oseltamivir who are 252 under age 25, between ages 25 and 45 inclusive and over age 45 are 4.138E-4, 1.379E-4, and 2.069E-253 4 respectively. Table 1 was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint 13 and the NAC over ten days. P(NA) on Day 0 is zero, however, as the days progress towards Day 10, it 262 approaches 1; on Day 10, it reaches 0.22 with a 5% and 95% confidence bounds of 0.2 and 0.25 263 respectively and a standard deviation of 0.42. Missing at least one pill (of the 10 pills) results in non-264 adherence. The results predict that 776 participants would take all their medications (10 pills) as 265 prescribed in ten days; 224 participants would miss at least one pill. This result aligns with the results 266 of the six studies: adherence to such treatment is very high. The NAF contributing the most to P(NA) 267 is the PatRel. Meaning, patient-related factors are strongly correlated to non-adherence. 268 to improve -that is, increase overall adherence rate. At a glance, it seems that patient-related factors 283 contribute the most to non-adherence. However, patient-related factors are not solely responsible for 284 non-adherence; other factors also contribute to non-adherence -this affirms results in previous studies 285 [1] . Further investigation of the results gives us a different picture. In Fig 5, it can be seen that patient-286 related factors contribute about 40% to the non-adherence probability. However, when the constituent 287 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. This study has established that NATA can reveal the non-adherence factors clinicians need to know to 293 allocate resources targeting those non-adherence factors. It is assumed that, given the information 294 produced by NATA, clinicians decide to reduce Forgot, Other, NoSym and SidEff by 20% each 295 through measures such as using a pillbox, software app, information/education [33] , trust in physician 296 [34] and psychological ownership [35] . The GoldSim model was updated and re-run to determine the 297 impact of the changes on non-adherence; the results are displayed in Fig 6. As expected, the overall 298 non-adherence of the improved system has reduced by nearly 4% at a mean of 0.187, 5% and 95% 299 confident intervals of 0.17 and 0.21 respectively and a standard deviation of 0.39. This reduced the 300 mean number of tablets wasted from 224 to 187 -saving 37 pills that could potentially increase the 301 evaluation of the efficacy of the treatment by 0.37%. 302 The significant changes made to the improved model using very generous reduction rates of 20% have 304 enhanced adherence by 3.7% -not as much as one would have expected. The reason for this big 305 change but relatively little impact is that all NAFs would have to be reduced to make significant 306 changes to the overall non-adherence. The results of this case study affirm that no single factor can 307 fully minimise non-adherence [4,11] and provides empirical proof. However, it is still clear that the 308 four main contributing NAFs are SidEff, Forgot, NoSym and Other; these are factors clinicians should 309 seek to improve to minimise non-adherence. 310 Adherence measuring techniques are usually implemented when a treatment has already begun. This 312 study has introduced NATA for predicting patients' adherence behaviour so that measures can be put 313 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint Adherence to long-term therapies: Evidence for action Temporal effectiveness of interventions to improve medication adherence: A network meta-353 analysis Adherence to medication Medication Adherence -Where Are We Today ? Overview The importance of adherence in tuberculosis 362 treatment clinical trials and its relevance in explanatory and pragmatic trials Economic 365 impact of medication non-adherence by disease groups: A systematic review Medication Adherence Mediates the Relationship Between Heart Failure 368 Symptoms and Cardiac Event-Free Survival in Patients With Heart Failure Interventional tools to 371 improve medication adherence: Review of literature Medication adherence: WHO cares? The challenge of patient adherence The Unmet Challenge of Medication Nonadherence Accuracy of a screening tool for 382 medication adherence: A systematic review and meta-analysis of the Adherence Scale-8 A new 385 taxonomy for describing and defining adherence to medications No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint Interventions for enhancing medication adherence Using 391 dispensing data to evaluate adherence implementation rates in community pharmacy Predictive validity of a medication 394 adherence measure in an outpatient setting Using the Medication Adherence Reasons Scale Scale) to identify the reasons for non-adherence across multiple disease conditions Handbook with Aerospace Applications. Washington DC: NASA Office of Safety and Mission 401 Assurance A Dyn Simul Approach to Reliab Model Risk Assess Using GoldSim The Novel Coronavirus -A Snapshot of Current Knowledge Remdesivir and chloroquine effectively 408 inhibit the recently emerged novel coronavirus (2019-nCoV) in vitro Medication adherence: The elephant in the room No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint Adherence with oseltamivir chemoprophylaxis among workers exposed to poultry during 414 avian influenza outbreaks in southern Israel Do general practice patients who are prescribed Tamiflu® actually 417 take it? Side effects of oseltamivir in end-stage 419 renal failure patients Recommendations for and compliance with social restrictions during implementation of school 423 closures in the early phase of the influenza A (H1N1) Adverse drug effects following oseltamivir 426 mass treatment and prophylaxis in a school outbreak of Compliance and side effects of prophylactic 430 oseltamivir treatment in a school in South West England Association of social support and 433 medication adherence in Chinese patients with type 2 diabetes mellitus Estimation of 436 the asymptomatic ratio of novel coronavirus infections (COVID-19) The authors would like to thank Prof. Peter Naude for reviewing the article and providing constructive 347 comments that have contributed immensely to its quality. 348 in place to improve adherence. NATA is not only a pre-treatment technique; it can also be used during 314 treatment. A NATA model in GoldSim can be updated with changes and re-run to determine how the 315 changes affect the system. This study has proven that NATA can identify non-adherence factors of a 316 treatment regimen and their relationships and contribution to overall non-adherence. With such 317 information, clinicians can implement mitigating strategies to minimise the risk of high non-318 adherence. Most of the data used in this study -extracted from other studies -are occurrence rates, 319 hence it was assumed they were exponentially distributed. However, the proposed solution, NATA 320 and GoldSim, are not restricted to exponentially distributed rates. Using Monte Carlo simulation, the 321 proposed solution can model and analyse any case study. 322 This study is not without limitations. The data used in the COVID-19 case study are based on a 324 similar drug -oseltamivir -of a similar disease. The authors assume that the behaviour of COVID-19 325 patients would be similar to that of the patients who took oseltamivir from six studies. The six studies 326 from which the data was extracted were diverse in terms of demographics and population; therefore, 327 for a geographically specific application, the data may need to be streamlined. The simulation for the 328 case study was modelled to run for ten consecutive days, which is not an accurate reflection of real-329 world studies where participants of a trial start on different days. NATA is not a stand-alone solution 330 for addressing all the issues with non-adherence; it depends on the results of studies and techniques 331 such as the Medication Adherence Reasons Scale or the Morisky Medication Adherence Scale 332 (MMAS) for data to perform its analysis. In the future, data for NAFs can be sourced from Big Data 333 and/or Artificial Intelligence-enabled systems where possible. 334 Non-adherence to a medication regimen is widespread. In addition to financial losses, non-adherence 336can blur the efficacy of drugs and lead to loss of lives. Most adherence measuring techniques are 337 implemented after the patient has started the medication regimen. This article has explored the use of 338Fault Tree Analysis (FTA) -an engineering technique for probabilistic risk analysis -to predict the 339 All rights reserved. No reuse allowed without permission.was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint (which this version posted July 3, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint All rights reserved. No reuse allowed without permission.was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint (which this version posted July 3, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint (which this version posted July 3, 2020. . https://doi.org/10.1101/2020.06.30.20135343 doi: medRxiv preprint