- Email: [email protected]
How to Measure ‘Opioid Relapse’ in Real-World Claims Data
A72
VA L U E I N H E A LT H 1 9 ( 2 0 1 6 ) A 1 - A 3 1 8
DPP-4 inhibitors are divided on 4 subgroups. The first two are patients on primary monotherapy with metformin and HbA1C level above 6.5%, and below 6.5%. The second two subgroups are patients on combination therapy and primary HbA1C level above 6.5%, and below 6.5%. A validated UNLOCK model was used to evaluate the risk for death and 7 most common diabetic complications development after 10 years. Results: GLP-1 agonists decreased the HbA1c level from 6.07% to 5.64% and from 8.45% to 7.53%, for the firsts and second subgroup respectively. The BMI decreased from 35.59 to 34.50 and from 37.42 to 36.15 in both sub groups. Patient’s weight is decrease from 100.91 to 99.29 and from 108.28 to 104.67. As a result, the cumulative risk for complications and death decreases with 9% and with 2% for the first subgroup, while the decrease for the second subgroup is 4% and 2% after a 10 years therapy. For patients treated with DPP-4 inhibitors and previous metformin therapy the initial HbA1C level decreased from 7.86% to 7.19%. Those on previous combination therapy decreased HbA1C to 7.94% for one-year period. Similar are the results for the other indicators. The cumulative risk for death decreased with 1% and risk of complication decreased with 2% to 4%. Conclusions: GLP-1 agonists and DPP-4 inhibitors benefit both the medical and social results of type 2 diabetes therapy by decreasing the risk of complications and death due to the diabetes. PRM5 A Review Evaluating the Validity of Smartphone Sensors and Components to Measure Clinical Outcomes in Clinical Research Byrom B1, Lee J2, McCarthy M3, Muehlhausen W3 Clinical Research, Marlow, UK, 2mProve Health, Arlington, VA, USA, 3ICON plc, Dublin 18, Ireland
1ICON
Objectives: Smartphones incorporate multiple inbuilt sensors to enhance user experience and provide certain features. More recently, novel application of these inbuilt sensors has enabled new and inventive uses for the smartphone in clinical research. We reviewed published validation studies performed that compare clinical outcomes assessments derived from smartphone sensor data to gold standard approaches. Methods: We categorised studies in our review into the areas of application and summarised validation findings. We eliminated studies validating PRO instruments and those using external sensors or adapters connected to smartphones. Results: Eight validation studies reported use of the smartphone accelerometer and gyroscope to measure joint angles and range of motion including shoulder, knee and spine, compared to goniometer, scoliometer and other approaches. Correlations varied between methods and outcomes measured: range of motion, flexion angles (r> 0.99), cervical rotation (r< 0.60). Balance and gait were assessed in five studies in comparison to research-grade accelerometers: showing moderate to strong correlations (r= 0.61 to 0.92) in displacement and time measures; two studies showed good correlation for some parameters measured in the Timed Up and Go test, and one in assessment of bradykinesia in Parkinson’s (PD) patients. One study showed good to strong correlation of tremor parameter measures via smartphone compared to ObsRO data (r= 0.69-0.87). Two studies showed high correlation between heart rate measured when the index finger is pressed against the rear camera and illuminated by the phone’s light-emitting diode in comparison to a pulse oximeter (ICC= 0.972, r= 0.99). Conclusions: Smartphone inbuilt sensors and components may offer a convenient and low cost approach to measurement of performance outcomes. Variability between approaches is inherent due to methodology and algorithm differences. Some approaches provide positive indications in comparison to gold standard methodologies, but more research in larger studies is encouraged for wider scale utility. PRM6 How to Measure ‘Opioid Relapse’ in Real-World Claims Data Montejano LB1, Ronquest NA2, Willson TM1, Wollschlaeger BA3, Cole AL1, Nadipelli VR2 Health Analytics, Ann Arbor, MI, USA, 2Indivior Inc., Richmond, VA, USA, 3Aventura Family Health Center, North Miami Beach, FL, USA
1Truven
Objectives: Opioid Use Disorder (OUD) is characterized by episodes of relapse and remission. Relapse is a key outcome and quality metric in OUD, but has no specific diagnosis code. This study examined different approaches to measuring relapse in claims data. Methods: Service-based indicators of opioid relapse were defined in several measures. The conservative measure (CM) included any of the following: diagnosis of continuous or episodic dependence following a dependence in remission diagnosis; hospitalization with a primary opioid-related diagnosis; detoxification with any opioid-related diagnosis; or emergency services with any opioid-related diagnosis. Inclusive measures (IMs) included the CM or any of the following indicators: hospitalization with a secondary diagnosis of opioid overdose; pharmacy claim for narcotic pain medications (NPMs) without indication of trauma/ surgery 7 days prior; pharmacy claim(s) for NPMs covering > 30 consecutive days; or abruptly stopping buprenorphine without taper. Relapse rates were measured among OUD patients initiating buprenorphine treatment in the MarketScan databases (2008-2014) and rates compared to the literature to assess measure feasibility. Results: Relapse rates 6 months post-treatment initiation ranged from 7.1% (CM) to 38.1% (IM) in Commercial patients (N= 22,563) and 6.0% (CM) to 44.0% (IM) in Medicaid patients (N= 7,811). Among the IM indicators, abrupt discontinuation was observed in 23.1% of Commercial patients and 26.6% of Medicaid patients. NPM claims without trauma/surgery were observed among 9.5% of Commercial and 16.7% of Medicaid patients. NPMs covering > 30 consecutive days were observed among 9.4% of Commercial and 8.8% of Medicaid patients. Conclusions: Relapse rates varied widely depending on the measure used. The literature similarly includes variation in rates due to measurement differences between studies. This highlights the need for a consistent definition of relapse in claims-based studies, the evidencebased utilization of defined measures, and future research targeting the validation of claims-based markers of opioid relapse. PRM7 Morbidity Endpoints (ME) in the Amnog Process in Germany: Are me Less Important in Oncology Substances?
Bakker K, Volmer T SmartStep Consulting GmbH, Hamburg, Germany
Objectives: The objective of this study is to assess the acceptance of morbidity endpoints by the Institute for Quality and Efficiency in Healthcare (IQWiG) within the early benefit assessment in Germany. The analysis focuses on morbidity endpoints (ME) for oncology substances compared to non-oncology substances: (1) how many ME have been accepted? (2) In how many cases has IQWiG determined an additional benefit based on ME? Methods: All benefit assessments published by the Federal Joint Committee (G-BA) between 01/01/2011 and 11/01/2016 have been considered (n= 158). Of those, 97 were excluded from the analysis due to one of the following reasons: orphan drug designation, no dossier submitted or incomplete evaluation by the IQWiG (due to missing data). Data analyzed: number of accepted ME as proportion of all submitted endpoints; additional benefit based on accepted ME; percent of assessments in which at least one ME was accepted. Results: In total, 24 oncology and 37 non-oncology assessments have been included into the analysis. For the 24 oncology assessments, a total of 101 ME had been submitted. IQWiG accepted hereof 34 ME (≈34%). Accepted ME include: pain, skeletal-related complications, health state and symptoms. The IQWiG determined an additional benefit for ten assessments based on ME (10/24≈42%). For the 37 non-oncology assessments, 163 ME had been submitted. IQWiG accepted 106 (≈65%) including: strokes, cardiovascular events and relapse-related events. For non-oncology substances, IQWiG determined an additional benefit for 19 assessments based on ME (19/37≈51%). At least one endpoint was accepted in 88% of oncology assessments and in 100% of benefit assessments in all other indications. Conclusions: Though the rate of accepting ME in oncology indications is numerically lower (34% vs. 65%), the derivation of an additional benefit based on ME in oncology assessments is comparable (43% vs. 51%). PRM8 Development of a Systematic Review Search Filter to Identify Medication Adherence Studies Johnson N1, Tongbram V1, Ndirangu K1, Ogden K2, Bay C3 Francisco, CA, USA, 3ICON, Cambridge, MA, USA
1ICON, New York, NY, USA, 2ICON, San
Objectives: Search filters or hedges are commonly used in systematic reviews to identify studies fulfilling specific criteria. Currently, there are no published filters that focus on identifying studies specific to patients’ medication-taking behavior, i.e. adherence, compliance or persistence. This research aims to develop and validate a medication adherence filter using schizophrenia as a case example. Methods: Published literature reviews assessing adherence in schizophrenia were identified from years 2000 to present in the English language, and were used as a reference. A filter search strategy was developed for adherence concepts using the titles of references identified in the reviews. This filter was used to search for medication adherence studies in schizophrenia in MEDLINE and Embase. This process was reiterated until studies included in the reference review articles were captured. Results: 21 published reference review articles were found on schizophrenia and adherence, which cited a total of 180 relevant studies. In the first iteration of the search strategy using the adherence filter, 142 studies were identified, thus capturing 79% of studies included in the reference reviews. After further revision and assessment of the missing articles, the final search filter was able to identify 89% of the reference articles. Conclusions: To our knowledge, this is the most comprehensive filter validation exercise for adherence at this time. The proposed filter was shown to be sensitive in the schizophrenia patient population and may be adapted to other disease areas after further testing. PRM9 Model Observational Bridging Study on the Effectiveness of Ezetimibe on Cardiovascular Outcomes Ferrières J1, Dallongeville J2, Getsios D3, Rossignol M4, Abenhaim L5, Grimaldi-Bensouda L6, Amzal B7 1Toulouse University, Toulouse, QC, Canada, 2Pasteur Institute, Lille, France, 3Evidera, Lexington, MA, USA, 4McGill University, Montreal, QC, Canada, 5LASER Analytica, London, UK, 6LASER Research, Paris, France, 7LASER Analytica, London, UK
Objectives: A Model-Observational Bridging Study (MOBS) integrated results from a population cohort of hypercholesterolemic individuals with predictive modelling to assess the public health impact of ezetimibe and estimate the number of cardiovascular events (CVE) potentially prevented by its use in real life. Methods: The 48-month prospective, nationwide cohort was conducted between 2008 and 2013 in France. Over 700 physicians recruited 3,395 patients with hypercholesterolemia started on ezetimibe as a lipid lowering therapy (LLT) initiation (15.2%), switch from another LLT (17.9%) or combined to a statin (63.9%). Patients were followed-up with annual physician visits and telephone interviews. MOBS leveraged cohort data to evaluate the effect of ezetimibe on cardiovascular morbidity and mortality over 5 years using longitudinal predictive modelling. Discrete event simulations used baseline and follow-up cohort data in addition to literature-based risk equations to estimate the number needed to treat (NNT) with ezetimibe to prevent a CVE. Model predictions were validated using actual number of CVE observed in the cohort. Results: Of a total of 9,314 person-years (p-y) of follow-up, 112 CVEs were observed: 96 non-fatal (myocardial infarction, stroke, acute coronary syndrome, coronary bypass surgery and angioplasty) and 16 fatal, for a rate of 12.02 events per 1,000 py (95% confidence interval (CI): 9.9 – 14.5) which was similar to that predicted by the model (11.95 per 1 000 p-y). The predicted reduction of CV events (non-fatal and fatal) for ezetimibe monotherapy as LLT initiators or switchers and ezetimibe combined to a statin were respectively 8, 2, and 12 per 1,000 patients treated over 5 years, with a global NNT of 143 patients over 5 years. These estimations were sensitive within 20% to the rate of treatment interruptions. Conclusions: MOBS methodology succeeded in efficiently combining predictive modelling and parsimonious data collection for robust public health impact assessment of drug utilization in real life.
VA L U E I N H E A LT H 1 9 ( 2 0 1 6 ) A 1 - A 3 1 8
DPP-4 inhibitors are divided on 4 subgroups. The first two are patients on primary monotherapy with metformin and HbA1C level above 6.5%, and below 6.5%. The second two subgroups are patients on combination therapy and primary HbA1C level above 6.5%, and below 6.5%. A validated UNLOCK model was used to evaluate the risk for death and 7 most common diabetic complications development after 10 years. Results: GLP-1 agonists decreased the HbA1c level from 6.07% to 5.64% and from 8.45% to 7.53%, for the firsts and second subgroup respectively. The BMI decreased from 35.59 to 34.50 and from 37.42 to 36.15 in both sub groups. Patient’s weight is decrease from 100.91 to 99.29 and from 108.28 to 104.67. As a result, the cumulative risk for complications and death decreases with 9% and with 2% for the first subgroup, while the decrease for the second subgroup is 4% and 2% after a 10 years therapy. For patients treated with DPP-4 inhibitors and previous metformin therapy the initial HbA1C level decreased from 7.86% to 7.19%. Those on previous combination therapy decreased HbA1C to 7.94% for one-year period. Similar are the results for the other indicators. The cumulative risk for death decreased with 1% and risk of complication decreased with 2% to 4%. Conclusions: GLP-1 agonists and DPP-4 inhibitors benefit both the medical and social results of type 2 diabetes therapy by decreasing the risk of complications and death due to the diabetes. PRM5 A Review Evaluating the Validity of Smartphone Sensors and Components to Measure Clinical Outcomes in Clinical Research Byrom B1, Lee J2, McCarthy M3, Muehlhausen W3 Clinical Research, Marlow, UK, 2mProve Health, Arlington, VA, USA, 3ICON plc, Dublin 18, Ireland
1ICON
Objectives: Smartphones incorporate multiple inbuilt sensors to enhance user experience and provide certain features. More recently, novel application of these inbuilt sensors has enabled new and inventive uses for the smartphone in clinical research. We reviewed published validation studies performed that compare clinical outcomes assessments derived from smartphone sensor data to gold standard approaches. Methods: We categorised studies in our review into the areas of application and summarised validation findings. We eliminated studies validating PRO instruments and those using external sensors or adapters connected to smartphones. Results: Eight validation studies reported use of the smartphone accelerometer and gyroscope to measure joint angles and range of motion including shoulder, knee and spine, compared to goniometer, scoliometer and other approaches. Correlations varied between methods and outcomes measured: range of motion, flexion angles (r> 0.99), cervical rotation (r< 0.60). Balance and gait were assessed in five studies in comparison to research-grade accelerometers: showing moderate to strong correlations (r= 0.61 to 0.92) in displacement and time measures; two studies showed good correlation for some parameters measured in the Timed Up and Go test, and one in assessment of bradykinesia in Parkinson’s (PD) patients. One study showed good to strong correlation of tremor parameter measures via smartphone compared to ObsRO data (r= 0.69-0.87). Two studies showed high correlation between heart rate measured when the index finger is pressed against the rear camera and illuminated by the phone’s light-emitting diode in comparison to a pulse oximeter (ICC= 0.972, r= 0.99). Conclusions: Smartphone inbuilt sensors and components may offer a convenient and low cost approach to measurement of performance outcomes. Variability between approaches is inherent due to methodology and algorithm differences. Some approaches provide positive indications in comparison to gold standard methodologies, but more research in larger studies is encouraged for wider scale utility. PRM6 How to Measure ‘Opioid Relapse’ in Real-World Claims Data Montejano LB1, Ronquest NA2, Willson TM1, Wollschlaeger BA3, Cole AL1, Nadipelli VR2 Health Analytics, Ann Arbor, MI, USA, 2Indivior Inc., Richmond, VA, USA, 3Aventura Family Health Center, North Miami Beach, FL, USA
1Truven
Objectives: Opioid Use Disorder (OUD) is characterized by episodes of relapse and remission. Relapse is a key outcome and quality metric in OUD, but has no specific diagnosis code. This study examined different approaches to measuring relapse in claims data. Methods: Service-based indicators of opioid relapse were defined in several measures. The conservative measure (CM) included any of the following: diagnosis of continuous or episodic dependence following a dependence in remission diagnosis; hospitalization with a primary opioid-related diagnosis; detoxification with any opioid-related diagnosis; or emergency services with any opioid-related diagnosis. Inclusive measures (IMs) included the CM or any of the following indicators: hospitalization with a secondary diagnosis of opioid overdose; pharmacy claim for narcotic pain medications (NPMs) without indication of trauma/ surgery 7 days prior; pharmacy claim(s) for NPMs covering > 30 consecutive days; or abruptly stopping buprenorphine without taper. Relapse rates were measured among OUD patients initiating buprenorphine treatment in the MarketScan databases (2008-2014) and rates compared to the literature to assess measure feasibility. Results: Relapse rates 6 months post-treatment initiation ranged from 7.1% (CM) to 38.1% (IM) in Commercial patients (N= 22,563) and 6.0% (CM) to 44.0% (IM) in Medicaid patients (N= 7,811). Among the IM indicators, abrupt discontinuation was observed in 23.1% of Commercial patients and 26.6% of Medicaid patients. NPM claims without trauma/surgery were observed among 9.5% of Commercial and 16.7% of Medicaid patients. NPMs covering > 30 consecutive days were observed among 9.4% of Commercial and 8.8% of Medicaid patients. Conclusions: Relapse rates varied widely depending on the measure used. The literature similarly includes variation in rates due to measurement differences between studies. This highlights the need for a consistent definition of relapse in claims-based studies, the evidencebased utilization of defined measures, and future research targeting the validation of claims-based markers of opioid relapse. PRM7 Morbidity Endpoints (ME) in the Amnog Process in Germany: Are me Less Important in Oncology Substances?
Bakker K, Volmer T SmartStep Consulting GmbH, Hamburg, Germany
Objectives: The objective of this study is to assess the acceptance of morbidity endpoints by the Institute for Quality and Efficiency in Healthcare (IQWiG) within the early benefit assessment in Germany. The analysis focuses on morbidity endpoints (ME) for oncology substances compared to non-oncology substances: (1) how many ME have been accepted? (2) In how many cases has IQWiG determined an additional benefit based on ME? Methods: All benefit assessments published by the Federal Joint Committee (G-BA) between 01/01/2011 and 11/01/2016 have been considered (n= 158). Of those, 97 were excluded from the analysis due to one of the following reasons: orphan drug designation, no dossier submitted or incomplete evaluation by the IQWiG (due to missing data). Data analyzed: number of accepted ME as proportion of all submitted endpoints; additional benefit based on accepted ME; percent of assessments in which at least one ME was accepted. Results: In total, 24 oncology and 37 non-oncology assessments have been included into the analysis. For the 24 oncology assessments, a total of 101 ME had been submitted. IQWiG accepted hereof 34 ME (≈34%). Accepted ME include: pain, skeletal-related complications, health state and symptoms. The IQWiG determined an additional benefit for ten assessments based on ME (10/24≈42%). For the 37 non-oncology assessments, 163 ME had been submitted. IQWiG accepted 106 (≈65%) including: strokes, cardiovascular events and relapse-related events. For non-oncology substances, IQWiG determined an additional benefit for 19 assessments based on ME (19/37≈51%). At least one endpoint was accepted in 88% of oncology assessments and in 100% of benefit assessments in all other indications. Conclusions: Though the rate of accepting ME in oncology indications is numerically lower (34% vs. 65%), the derivation of an additional benefit based on ME in oncology assessments is comparable (43% vs. 51%). PRM8 Development of a Systematic Review Search Filter to Identify Medication Adherence Studies Johnson N1, Tongbram V1, Ndirangu K1, Ogden K2, Bay C3 Francisco, CA, USA, 3ICON, Cambridge, MA, USA
1ICON, New York, NY, USA, 2ICON, San
Objectives: Search filters or hedges are commonly used in systematic reviews to identify studies fulfilling specific criteria. Currently, there are no published filters that focus on identifying studies specific to patients’ medication-taking behavior, i.e. adherence, compliance or persistence. This research aims to develop and validate a medication adherence filter using schizophrenia as a case example. Methods: Published literature reviews assessing adherence in schizophrenia were identified from years 2000 to present in the English language, and were used as a reference. A filter search strategy was developed for adherence concepts using the titles of references identified in the reviews. This filter was used to search for medication adherence studies in schizophrenia in MEDLINE and Embase. This process was reiterated until studies included in the reference review articles were captured. Results: 21 published reference review articles were found on schizophrenia and adherence, which cited a total of 180 relevant studies. In the first iteration of the search strategy using the adherence filter, 142 studies were identified, thus capturing 79% of studies included in the reference reviews. After further revision and assessment of the missing articles, the final search filter was able to identify 89% of the reference articles. Conclusions: To our knowledge, this is the most comprehensive filter validation exercise for adherence at this time. The proposed filter was shown to be sensitive in the schizophrenia patient population and may be adapted to other disease areas after further testing. PRM9 Model Observational Bridging Study on the Effectiveness of Ezetimibe on Cardiovascular Outcomes Ferrières J1, Dallongeville J2, Getsios D3, Rossignol M4, Abenhaim L5, Grimaldi-Bensouda L6, Amzal B7 1Toulouse University, Toulouse, QC, Canada, 2Pasteur Institute, Lille, France, 3Evidera, Lexington, MA, USA, 4McGill University, Montreal, QC, Canada, 5LASER Analytica, London, UK, 6LASER Research, Paris, France, 7LASER Analytica, London, UK
Objectives: A Model-Observational Bridging Study (MOBS) integrated results from a population cohort of hypercholesterolemic individuals with predictive modelling to assess the public health impact of ezetimibe and estimate the number of cardiovascular events (CVE) potentially prevented by its use in real life. Methods: The 48-month prospective, nationwide cohort was conducted between 2008 and 2013 in France. Over 700 physicians recruited 3,395 patients with hypercholesterolemia started on ezetimibe as a lipid lowering therapy (LLT) initiation (15.2%), switch from another LLT (17.9%) or combined to a statin (63.9%). Patients were followed-up with annual physician visits and telephone interviews. MOBS leveraged cohort data to evaluate the effect of ezetimibe on cardiovascular morbidity and mortality over 5 years using longitudinal predictive modelling. Discrete event simulations used baseline and follow-up cohort data in addition to literature-based risk equations to estimate the number needed to treat (NNT) with ezetimibe to prevent a CVE. Model predictions were validated using actual number of CVE observed in the cohort. Results: Of a total of 9,314 person-years (p-y) of follow-up, 112 CVEs were observed: 96 non-fatal (myocardial infarction, stroke, acute coronary syndrome, coronary bypass surgery and angioplasty) and 16 fatal, for a rate of 12.02 events per 1,000 py (95% confidence interval (CI): 9.9 – 14.5) which was similar to that predicted by the model (11.95 per 1 000 p-y). The predicted reduction of CV events (non-fatal and fatal) for ezetimibe monotherapy as LLT initiators or switchers and ezetimibe combined to a statin were respectively 8, 2, and 12 per 1,000 patients treated over 5 years, with a global NNT of 143 patients over 5 years. These estimations were sensitive within 20% to the rate of treatment interruptions. Conclusions: MOBS methodology succeeded in efficiently combining predictive modelling and parsimonious data collection for robust public health impact assessment of drug utilization in real life.