x
Kenkyu Journal of Pharmacy Practice & Health Care ISSN : 2455-4421
Establishing the influence that Polypharmacy has on Patient well-being
  • Sparks R* ,

    CSIRO, Digital Flagship, Sydney, Australia, E-mail; Ross.Sparks@csiro.au

  • Okugami C ,

    CSIRO, Digital Flagship, Sydney, Australia

  • Dearin J ,

    University of Notre Dame Medical School, Sydney, Australia

Received: 14-08-2015

Accepted: 24-08-2015

Published: 31-08-2015

Citation: Sparks R (2015) Establishing the influence that Polypharmacy has on Patient well-being. Pharma Health Care 1: 100101

Copyrights: © 2015 Sparks R

Abstract

Medication is one of the most frequent ways of managing patients suffering from a chronic disease. Many chronically sick patients use a wide range of drugs to improve their well-being.  However, often the interaction between the cocktails of prescribed medications is not well understood.  This paper investigates the impact (adverse or positive) drug reactions, involving multiple medications, have on patient’s well-being.  The aim is to detect sudden changes in health condition using machine-learning techniques.   This technology examines the impact multiple drugs have on patient’s perceived well-being.  

 

Keywords: Polypharmacy; Monitoring adverse drug reactions.

Introduction

This paper explores the relationship between the patient changes in prescription medication profiles and the patient’s self-assessment on how well they feel. The main aim is to try and assess the impact drug changes have on a patient’s perceived wellness. It also facilitates the patient self-assessment and helps the patient manage their lifestyle risks (e.g., all indicators point to me not going out today because many of my measurements are negative).


The data source of information is from a trial involving 100 to 125 test patients who had a tele-monitoring device installed in their home for monitoring their vital sign measures. [1,2] For the duration of the study we have collected information on prescription medication purchases for the test patients as well as monitored their perceived wellness on a daily basis. Wellness measures resulted from responses to daily survey questions on how well patient feel.


The monitoring of the vital signs data has been covered in. [1] Please note that the selected test patients involve a cohort of chronically sick patients who are 55 or over in age, and who are not currently cancer suffers. Some have been monitored for over 18 months but all more than 6 months


These vital signs data are not being explored fully in this paper because the focus is on a wellness variables resulting from daily questionnaire data collected on whether the patient feels better than usual or worse than usual.
The link between wellness and the total cost of medication is outlined in Figure 1 for a particular patient in the study cohort. Notice that when the patient perceived wellness tracks lower in 2013 the drug total costs tracked higher and thereafter when medication cost went down well-being improved. The data in Figure 1 is simultaneous so it is difficult to separate cause from effect. All the graph indicates is a correlation. The interesting question is whether medication costs can be controlled by improving the patient perceived wellness?

 

 

Figure 1: Cost of patient drugs influence on individual patient’s wellness. (assessment. 0 – feeling the same as usual, -1 feeling worse than usual, -2 feeling terrible, 1 feeling better than usual, 2 feeling terrific .)

 

This counter the placebo effect where we expect patient to feel better once they are medicated, and so the interesting question is what influence do drugs have on perceived well-being and can we establish the adverse drug or positive drug influence for particular patients very early and improve outcomes related to medicating patients? Although we are unlikely to answer this question with this research, but we hope to convince readers that this maybe an important question worth answering.


The methodology of flagging positive and adverse drug influences on individual patients is outlined in section 2 in terms of how they influence patient perceived well-being. Section 3 discusses a few adverse drug interactions that influence how well particular patients feel in the study cohort. Section 4 closes with some conclusions on how general practitioners could use the methodology to improve patient perceived well-being. All patients in our study cohort and chronically sick patients were studied with the aim to improve the rest of their life by the appropriate management of their medication. This fits well with the recent proposals to include non-dispensing pharmacists in General Practice to improve medication compliance and reduce adverse medication outcomes.

 

Methodology for providing evidence of drug influence on well-being


We only have the medication benefit payment recorded amount spent on a particular drug and we use the total price paid for the same drug as the surrogate for dose. Although we could assume that the more spent on the drug the higher the dose this assumption was not made and the cost information was not used. In addition, we assume that a patient purchases a drug to-day then they are likely to start using it immediately and its influence on the patient’s well-being will last for the next two weeks. The patient’s daily well-being measures are denoted wt for day t, and are defined by


wt = 2 if the patient feels terrific;
wt = 1 if the patient feels better than usual;
wt = 0 if the patient feels the same as usual;
wt = -1 if the patient feels worse than usual;
wt = -2  if the patient feels terrible;


 
If two weeks before the patient spent zero on a particular drug then they are assumed not to be on that drug and its influence on the patient is assumed to be zero. However if they purchased a particular drug at time t-k (k<14) then we assumed the drug is taken from that day and it influences well-being immediately for the next two weeks. A zero-one variable is used to indicate whether the drug is influencing a patient’s well-being; a zero when the patient is assumed to be not on the drug for the past two weeks and a one when the patient is assumed to be on the drug. If the person is on several other drugs we repeat the process now using another variable for that drug. These values are now selected as explanatory values for forecasting the well-being of patients using a regression tree model. We could use classification trees rather than regression trees (for continuous variables) but we found that the nurses we were working with understood regression trees better.


These regression models are used to define what combination of drugs and “dose levels” make them feel better and what combination make them feel worse than usual. The regression tree models performed well particularly when there are complicated interactions (i.e. influence of using drug A&B interaction). This is exactly the case that we are trying to establish when combinations of drugs have a positive or a negative influence on a patient’s perceived well-being, modelling technique can be simple as taking wt as the response variables and the drug indicator values for the patient as the explanatory variables.


Figure 2 outlines the goodness of fit for such models for patient 1 in our study. First it is easy to see that this patient seldom feels better than usual and this generally occurs early in the study, thus indicating that this patient is probably deteriorating in health over the study period. This plot indicates that the model fits reasonable well (explains 90% of the variation) especially due to the fact that this model is forecasting one day ahead.

 

Figure 2: Patient 1’s wellbeing measures and the regression tree model fits.

Results

 

The first step in monitoring the wellness values wt is to assess whether the wellness values are predictable and therefore in-control. Here we assess whether the model is adequate at describing either adverse drug reactions or unusually positive well-being outcomes. Table 1 below examines the model mean square error (MSE) for the following situation; using:


 wt = 0 to predict how well the patient feels to-morrow (the baseline situation where no information is used by nurses to predict the patient wellbeing– this is the current situation); 


wt given we know the value of wt-1 using regression trees to fit the data,i.e., using the to-days patient’s well-being to predict this patient’s to-morrows wellbeing; 


wt given we know the value of wt-1 and past vital signs measures using regression, i.e., this is the approach proposed in this paper using to-days patient’s well-being measure and their past vital signs measures to predict their tomorrow’s wellbeing.

 

 

 

Table 1: The MSE (mean square errors) for the various approaches to predicting the patient’s wellbeing one day ahead.

 

Table 1 indicates that to-days patient 1’s well-being measure is relative more useful at predicting this patient’s to-morrows well-being measure than for patients 14 and 77. In other words, the proportion of information explained by wt-1 is (0.2461-0.1567)/ 0.2461=0.36 for patient 1while for patient 14 and 77 it is only 0.02 and 0.20. Using the additional information of vital signs the proportion of variation explained for patients 1, 14 and 77 is 0.58[= (0.2461-0.1027)/ 0.2461], 0.43 and 0.52, respectively.


Figure 2 illustrated the model goodness of fit graphically using the same data that fitted the model. Next we examine the moving variable differences between the day-ahead forecast errors to assess when the model is forecasting accurately and when it is forecasting less well. Let us denote the model forecast for the day ahead made on day t as 

 

 

Then the moving estimate of the one day ahead forecast error is given by

 

Where some starting value for the variance, is generally established using historical data and  is the exponential weight which is used to estimate the local uncertainty. The z-score is defined by    which is used to establish when the model forecasts are out-of-control using 3 sigma control charts (Shewhart charts). The last plot in Figure 3 is the plot of  and the changes in  assessed by rsp package [3]. Change points in the uncertainty and are flagged using the rsp package. [3] These are in blue in the first and third plot of Figure 3 respectively, and they indicate that the model prediction uncertainty does change significantly and the level of wellness does change significantly during the study period. The second plot in Figure 3 indicates that the forecasts are generally in-control, e.g., well-being is predictable. For the third plot in Figure 3, the blue line is the rsp value of the estimated change point in well-being and the green line is the actual value. In the third plot, the red line and blue line provide information on trends and change points in wellbeing.

Figure 3: Assessing the model and measurement uncertainty.

 

Now we review the well-being for select few- patients for the duration of the study periods and flag potential drug influences on well-being. It is important to recognise that this monitoring tries to establish the influence of comfort medication (such as morphine, anti-depressant etc. influenced to the wellness assessment answer of patient) and therefore we should avoid any inferences made about the long-term influences of drugs in this monitoring approach.

 

Patient 1

 

This patient has two co-morbidities: cardiovascular disease and chronic obstructive pulmonary disease. This is a common disease combination in Australia especially for elderly patients. The majority of sufferers are being medicated for both of these conditions. This patient’s doctor has the challenge of managing the patient’s health risks (physical) and well-being in terms of quality of life and mental state.

 

Figure 4: Recursive tree regressive model describes the influences of wellness.

 

The terminal nodes of the regression tree in Figure 4 are used to predict how well the patient is going to feel given a set of rules for the past 7 days vital sign measurements, e.g., for the node on the far left (node32) of Figure 3 the predicted well-being measure is -0.98 (patient feels worse than usual). This occurs generally when the peak expiratory flow (pef) measure yesterday is less than 350L/min, heart rate one day ago was less than 97 beats per minute, and forced expiratory vital capacity (fvc) five days ago is less than 2.3L. On the other hand, the patient will generally feel better than usual if the pef measure yesterday was higher than 500L/min and body temperature is above 37deg.C(node:3). The closer the terminal node is to the color green in the tree the closer the tree model predicts that the patient is going to feel better than usual, in contrast red cells predict that the patient will feel unwell. Figure 3 and Figure 4 indicates that the vital signs data is very useful in predicting how well the patient is going to be a day ahead, and therefore this could be used to regulate the care program for the patient.


The advantage of the information in Figure 4 is that normal pef varies from individual-to-individual based on their age, height, gender, etc. and so it is hard to judge which value describes the patient’s normal well-being level. Figure 4 provides help with establishing an answer (e.g., greater than 5).


Now we use the same tree modeling approach to discover the influence drugs have on well-being. We assume that drugs purchased are consumed within the two weeks after purchased. We use this medication purchase data to understand the change and influence of how well the patient feels during those two weeks.


The drug takes on a value of 1 if it had been purchased within the past two weeks, and 0 otherwise. The aim here is to determine the influence the drug has on how well the patient feels, either positive or negative. In Figure 5, if the drug is greater than 0.5 then it is assumed that the patient is taking the drug, otherwise patient 1 is not taking it.

The results from Figure 5 are as follows:

 

If the patient takes Ipratropium but no Phenobarbitone and Ceftriaxone then the patient is likely to feel unwell.

If the patient takes Ipratropium and Phenobarbitone then the patient is unlikely to feel unwell.

If the patient does not take any medication then they are unlikely to feel unwell.

If the patient takes Gentamicin and Itraconazone then they are unlikely to feel unwell.

None of this cocktail of drugs seem to have an immediate positive influence on the patient’s well-being by making them feel better than usual.

 

Figure 5: The regression tree for establishing the influence drug combination has on wellbeing.

 

Negative influence of taking Ipratropium: Ipratropium is a drug that relieves bronchial spasm. It is an anticholinergic drug used for the treatment of chronic obstructive pulmonary disease and asthma. [4] This drug has no significant influence on increased mortality.[5]There are a number of adverse effects associated with ipratropium including headache, nausea and taste disturbances (>1%), [6] Ipratropium  appears to decrease perceived wellness for this patient. Significant decline in wellness can be seen in Figure 6a on two occasions when the drug was started, but on two other occasions there were no adverse influences. Therefore the adverse influence of this drug is in doubt.

 

   

Figure 6a: Patient 1’s wellness trends in relation to when assumed to be on Ipratropium.

 

Negative influence of Ipratropium and/or Gentamicin: Gentamicin is used to treat Gram negative bacterial infections. Gentamicin is toxic and dose control is critical. From Figure 6 it looks as if when these drugs are taken simultaneously then the patient feels unwell, however the evidence this is not convincing.

 

 

Figure 6b: Patient 1’s wellness trends in relation to when assumed to be on Ipratropium and/or Negative influence of Ipratropium and/or Itraconazole: Itraconazole is used to treat fungal infections in the lungs. There is no strong evidence of an adverse influence from taking these drugs together (see Figure 6c) but some studies report problems with this drug. [7]

 

 

Figure 6c: Patient 1’s wellness trends in relation to when assumed to be on Ipratropium and/or Itraconazole.

 

Positive influence of taking Phenobarbitone and Ipratropium: Taking a combination of these medications appears to have a positive influence on wellness since on the occasions when these were taken on their own then there was a smaller tendency to feel unwell but when Ipratropium was taken with Phenobarbitone then the patient appeared to feel less unwell. Note however that Phenobarbine is known to have long-term adverse effects. [8]

 

 

 

 

Figure 6d: Patient 1’s wellness trends in relation to when assumed to be on Ipratropium and/or Phenobarbitone. Such information may be very useful for the GP to observe and use to regulate the patient medication in a way that improves the chance of the patient feeling well. There is a danger with the focus being short term rather than long-term. However, long-term influences are harder to establish because of the number of confounding factors.

 

Patient 77

Based on the medication this patient is taking; this patient has a cardiovascular problem and kidney failure. The patient is also an asthma sufferer. The patient’s difficulty with breathing can be observed by monitoring the PEF vital sign measurement over time. The increase in pef score is evident after a bushfire event (Figure 7 below).

Figure 7: Impact of local bushfire on patient 77’s breathing ability.

 

In Figure 8 we establish the past vital sign measures that impact on patient 77’s .wellness scores. If this patient’s systolic blood pressure (6 days before) exceeds 156 and diastolic blood pressure (the day before) exceeds 68 then the patient is more likely to feel worse than usual. The heart failure and difficulty with breathing for patient 77 is shown by the number of times SpO2, FEV1, FVC, heart rate (egc) and blood pressure are partitioned on in Figure 8.

 

Figure 8: Forecast model for patient 77’s well-being.

 

For this patient, the difficulty with breathing can be treated using Ipratropium by decreasing cholinergically mediated bronchospasm thus providing higher intake of air and consequent tissue oxygenation. Also Fluticasone, an inhaled topical corticosteroid is used for relieving difficulties with breathing and chest tightness in asthma sufferers by reducing bronchial inflammation, thus helping to improve the overall wellness. Increasing oxygen intake reduces the risk of heart failure. [9]

 

The influence of “past” medication on patient 77’s well-being is clear in Figure 9 where we examine the influence drug intake has on patient 77’s wellbeing. If the patient takes omeprazole and does not takes ipratropium or fluticasone then the patient is likely to feel worse than usual. The patient feels better than usual when they are not taking omeprazole. If the patient takes omeprazole then they should simultaneously take ipratropium to not feel unwell.

 

Figure 9: Medication’s influence on well-being.

 

Omeprazole belongs to group of drugs called proton pump inhibitors. It decreases the amount of acid produced in the stomach. Omeprazole is well tolerated, highly effective, and safe for treatment of erosive esophagitis and symptoms of gastroesophageal reflux in children and adults. [10] However this patients seem to be adversely influenced by taking this drug in Figure 10a because after or during taking this drug the patient well-being assessment is lowered. This potential adverse influence should be monitored more closely in future.


Figure 10a: The influence of taking omeprazole on patient 77’s well-being.


 

Positive impacts of medication

Patients with dyspnoea secondary to COPD or asthma can be treated using Ipratropium. Also Fluticasone oral inhalation is used for preventing breathing difficulties and chest tightness. In addition, there is some concern in the literature for asthma sufferers with using this drug.[11-15] Inhaled corticosteroids (ICS) may increase the risk of pneumonia and in large doses cause adrenal suppression, osteoporosis, thinning and bruising of the skin and increased risk of glaucoma.(AMH)


However increasing oxygen intake reduces the risk of heart failure, and Figure 10b shows some evidence of improved well-being by taking Ipratropium on its own and Ipratropium and Fluticasone simultaneously. This is not proof but it does indicate that some patient’s responses to drugs could be counter to conventional wisdom. This highlights the importance of monitoring individual patient’s response to drugs however the long-term influences could be different and this needs to be always considered in the risks different drugs pose to patients.

Patient 14

Patient 14 is taking ciprofloxacin (an antibiotic), frusemide (cardio-vascular drug – a loop diuretic), oxycodone (an opioid) and prazosin (cardio-vascular drug – an alpha blocker). The patient does not feel well when taking ciprofloxacin on its own, but if the patient takes this drug with oxycodone then the patient feels much better. This could be because of the relative pain relief obtained by taking an opioid. If patient 14 takes oxycodone and frusemide then they don’t feel that well.

 

Conclusion

Combination of data analytics and daily measurements of telehealth system, successfully detects the complex pattern of medication influence over time of our test patients who have variety of different conditions.

Acknowledgement

The authors wish to acknowledge the Australian Government for its funding of this project through its Telehealth Pilots Program and CSIRO eHealth Research Program.

Reference

  1. Celler BG, Sparks RS (2015) Home telemonitoring of vital signs – Technical Challenges and Future directions.  IEEE Journal of Biomedical and Health Informatics 19: 82-91.

  2. Celler BG, Sparks R, Nepal S, Alem L, Varnfield M, et al. (2014) Design of a multi-site multi-stage clinical trial of home monitoring of chronic disease in community in Australia. BMC Public Health 14:1270.

  3. Capizzi G, Masarotto G (2013) Phase I Distribution-Free Analysis of Univariate Data.  Journal of Quality Technology  45: 273-284.

  4. Sin DD, Tu JV (2000) Lack of association between ipratropium bromide and mortality in elderly patients with chronic obstructive airway disease. Thorax 55: 194-197.

  5. Ringbaek T, Viskum K (2003) Is there any association between inhaled ipratropium and mortality in patients with COPD and asthma?. Respiratory medicine  97: 264-272.

  6. Australian Medicines Handbook 2015. PO Box 240 Rundle Mall, Adelaide SA 500.

  7. Pohjola-Sintonen S, Viitasalo M, Toivonen L, Neuvonen P (1993) Itraconazole prevents terfenadine metabolism and increases risk of torsades de pointes ventricular tachycardia. European journal of clinical pharmacology 45: 191-193.

  8. Rzany B, Correia O, Kelly JP, Naldi L, Auquier A, et al. (1999) Risk of Stevens-Johnson syndrome and toxic epider mal necrolysis during first weeks of antiepileptic therapy: a case-control study. The lancet, 353: 2190-2194.

  9. Wilson J, Woods I, Fawcett J, Whall R, Dibb W, et al. (1999) Reducing the risk of major elective surgery: randomised controlled trial of preoperative optimisation of oxygen delivery. BMJ 318:1099-1103.

  10. Hassall E, Israel D, Shepherd R, Radke M, Dalväg A, et al. (2000) Omeprazole for treatment of chronic erosive esophagitis in children: a multicenter study of efficacy, safety, tolerability and dose requirements. The Journal of pediatrics 137: 800-807.

  11. Teper AM, Kofman CD, Szulman GA, Vidaurreta SM, Maffey AF (2005) Fluticasone improves pulmonary function in children under 2 years old with risk factors for asthma.Am J Respir Crit Care Med  15;171:587-590. 

  12. Orsini L, Limpa-Amara S, Crown WH, Stanford RH, Kamal K (2004) Asthma hospitalization risk and costs for patients treated with fluticasone propionate vs montelukast. Annals of Allergy, Asthma & Immunology  92: 523-529.

  13. Grounds RO, Bennett E. A Randomized Clinical Trial of the Effect of Deliberate Perioperative Increase of Oxygen Delivery on Mortality in High-Risk Surgical Patients. JAMA.

  14. Capizzi G, Masarotto G (2012) Adaptive generalized likelihood ratio control charts for detecting unknown patterned mean shifts. Journal of Quality Technology 44, 281–303.

  15. Whysner J, Ross PM, Williams GM (1996) Phenobarbital mechanistic data and risk assessment: enzyme induction, enhanced cell proliferation, and tumor promotion. Pharmacology & therapeutics, 71(1), 153-191.

Signup to recive email updatesx