This is the first study to assess the effects of changing SMBG practice on glycemic control over a longer period and to separately evaluate new and ongoing SMBG users. We observed a significantly different effect between new and ongoing users, suggesting that pooling may have biased previous analysis. Among new users, initiating SMBG was associated with a graded improvement in glycemic control, even among those not treated pharmacologically. Among prevalent users, there was a significant association between change in SMBG and A1C only in those receiving pharmacologic therapy; decreases in SMBG frequency were significantly associated with a modest worsening in glycemic control, whereas increases in SMBG were associated with modest improvements in control. All of the above associations were significant and graded but with diminishing returns.
Among new users, the biological feedback associated with SMBG may help patients understand how behaviors (e.g., exercise and diet) and clinical states (e.g., different insulin doses or timing) impact their glycemic control. Among prevalent users not receiving pharmacologic therapy, changes in SMBG had no substantive impact on A1C; perhaps this constant feedback maintains value only if used in real time to guide behavioral and therapeutic modifications or recorded to inform their provider about glycemic patterns.
In prevalent users, the impact on control of changing average testing frequency by one test per day was somewhat smaller in insulin-treated than in OHA-only users (0.12- vs. 0.16-point decrease in A1C for a one strip per day increase in monitoring frequency; P
< 0.0001), perhaps because insulin-treated patients typically require multiple tests throughout the day to adjust insulin, whereas OHA-only patients rely on less frequent testing. It has been demonstrated that well-informed insulin-treated patients readily modify insulin dose and timing in response to SMBG readings and that improved insulin administration effectively improves glycemic levels (11
). Patients treated by OHAs only are rarely instructed to change or titrate dose themselves in response to home glucose readings; however, they may change timing of oral medications. Our findings differ from a study concluded by the Veterans Administration (12
). These authors reported no significant worsening in A1C after modest reductions in SMBG frequency (reduction of 0.4 and 0.6 strips/day on average) among stable, diet-controlled type 2 diabetic and oral agent–treated patients, respectively, all of whom were prevalent SMBG users at baseline. It is important to note that after the reduction, both groups were still testing quite regularly (~0.7 strips/day). It is unclear as to whether SMBG lacked effect or, alternatively, whether their patient population was overutilizing SMBG prior to their study, and the reduction in SMBG frequency was small enough to cross a threshold after which further reductions in utilization would have caused harm.
Several potential limitations and strengths of this study deserve comment. Causal interpretation of these findings is limited by the lack of randomization. However, randomized controlled trials also have limitations for the study of SMBG. Blinding is not possible in studies of SMBG, and thus such studies are potentially biased when patient preferences are not incorporated (16
). For such interventions that are not amenable to blinding, a strong patient preference and randomization to the nonpreferred intervention can reduce the internal validity of an randomized controlled trial by biasing the estimate of effect (16
). Randomized controlled trials may be biased when behavioral improvements are stimulated by the subjects' knowledge that their outcomes will be observed (Hawthorne effect) (13
). Furthermore, most current SMBG practice guidelines make it unethical to include an unexposed arm. Thus, we suggest that these observational findings are an important complement to those derived from experimental studies.
Observational studies of SMBG are susceptible to well-described biases, which require analytic attention. Reverse causality (endogeneity) is one potential source of bias in observational studies if not analyzed correctly. This would occur if changes in the study outcome (glycemic control) actually led to subsequent changes in the exposure (SMBG), instead of the other way around. We found some evidence of such a pattern. We evaluated glycemic control among subjects not using SMBG (i.e., before initiation of SMBG in the new-user cohort) and found that initiators of SMBG had substantially poorer control versus nonusers who did not initiate SMBG during the follow-up (A1C 9.3 vs. 8.2% in insulin-treated, 8.6 vs. 7.3% in OHA-only, and 8.2 vs. 6.6% in diet-controlled patients).
Another limitation is that patient receipt of education or instruction on SMBG practice is unknown to us. Kaiser Permanente offers diabetes health education classes that include training on the use of SMBG when diabetes is initially diagnosed and further on in the natural history of diabetes; Kaiser Permanente SMBG guidelines are in line with the American Diabetes Association clinical recommendations for SMBG. In accord with those guidelines, provider recommendations are flexible and will likely vary over time, depending on changes in type of therapy, disease severity, patient motivation, etc. We do not know how well health educators or care providers follow guidelines, to what extent provider recommendations change over time, how providers instruct the patients to use the SMBG data, or how well patients understand or incorporate these instructions.
Poor glycemic control likely motivates health care providers to urge their patient to initiate or intensify SMBG practice or may motivate the patient to initiate on his/her own. A simple cross-sectional anlysis of this new-user cohort would have resulted in distorted (and counterintuitive) findings, suggesting that SMBG initiators had 1- to 2-point–poorer glycemic control, i.e., the opposite direction of effect observed in our longitudinal analysis of new users. This may account for some of the negative findings in previous observational studies that did not separate new and prevalent users. Another possible source of endogeneity would be that providers could simultaneously intensify diabetes therapy because of poor control and recommend increasing SMBG. Thus, the improvement in A1C could be due to therapy intensification, SMBG, or both. Since our intention was to evaluate the SMBG effect, we therefore excluded any subjects that initiated new therapies during follow-up subsequent to the baseline A1C measurement. However, our study is limited by the lack of data on change in medication dose. Thus, the limits of our data also preclude us from separating the impact of SMBG results influencing providers to modify the patient's therapy dose from the impact of SMBG on self-care. However, we see both pathways as potentially important benefits of SMBG, and the benefit likely differs on an individual basis; therefore, quantifying the importance of each pathway may be of secondary importance from a public health point of view.
We further addressed these concerns of endogeneity by specifying an augmented regression (17
) to test the null hypothesis that SMBG is exogenous (predetermined) in the A1C regression. This method is widely used in the econometrics literature where observational data analyses are the norm. In brief, the augmented regression is equivalent to putting the predicted values of the potentially endogenous regressor (i.e., change in SMBG) into the second-stage model for the outcome (change in A1C), along with the residuals from the first-stage prediction model. The prediction model includes an “instrumental variable;” in this case, an exogenous policy change that altered the copayment for test strips but should not have had an impact on A1C independent of its effect mediated through SMBG. Failure to reject the null hypothesis that the residuals have any effect implies that changes in SMBG utilization are exogenous and that the single-equation estimates are consistent, i.e., two-stage instrumental variable modeling is unnecessary. Intuitively, the effect of the predicted regressor is the instrumental variables estimate of the unbiased “causal” effect of that regressor. Therefore, if the residuals from the prediction model for SMBG demonstrate independent explanatory power in the A1C regression, it must be because the error terms in the SMBG and A1C equations are correlated, i.e., SMBG is endogenous to A1C. As this was not the case with our data, we present only the single-equation estimates. It is worth noting that the most likely form of endogeneity (poorer glycemic control leading to increased monitoring) would have biased findings toward the null, and thus, if anything, our findings should be conservative. Omitted-variable bias is another limitation of observational study design. For example, SMBG frequency may simply be a marker for better, more intensive disease management, which more directly influences health outcomes. Because this study was conducted in a single health maintenance organization, the quality of diabetes care was probably more uniform than in a community sample. However, differences in disease management approach across providers and over time could confound our findings. Therefore, we assessed the sensitivity of our regression results by simulating how strong an unmeasured (residual) confounder would have to be to make our findings attenuate to nonsignificance (18
). This analysis showed that only a very potent residual confounder would negate our conclusions. Such an unmeasured confounder would have to confer an effect comparable to initiating a pharmacological agent (e.g., 1% lowering of A1C) and be highly prevalent in those who practice SMBG and not prevalent in those who do not practice. While theoretically possible, we think it is unlikely such a strong confounder (associated with SMBG and control) would have gone unnoticed.
SMBG utilization may have been slightly underascertained if members purchased additional testing supplies in non–Kaiser Permanente pharmacies, although exclusion of the ~5% of patients lacking pharmacy benefits should have minimized this potential misclassification. The exclusion of individuals who change diabetes therapy reduces the generalizability of this study. However, we felt that this exclusion was necessary to avoid the observed change in A1C being more due to change in therapy than due to change in SMBG (i.e., a threat to internal validity). Exclusion of those individuals who, on the basis of SMBG, advanced their pharmacologic therapies may have caused us to underestimate the potential benefit of such an intervention.
A unique strength of this study is its longitudinal design and duration of follow-up (4 years). The longest study included in the most recent systematic review by Welschen et al. (13
) was <1 year, while the remainder of the studies were concluded within 6 months. Another strength of this study is the distinct new-user and ongoing-user design, which carefully accounts for the timing of SMBG initiation. The comprehensive data included in the Kaiser Permanente Northern California Diabetes Registry make it possible to adjust for important potential confounders, including diabetes self-care practices, medication adherence, and lifestyle behaviors, each of which may be independently associated with monitoring frequency and glycemia. We have shown previously that self-care and health behaviors (e.g., better refill adherence and lower rates of smoking) and appropriate annual screenings were more common in patients who adhered to the American Diabetes Association SMBG guidelines (7
). Thus, self-monitoring could simply be a marker for more intensive disease management or better self-care, which more directly impacts glycemia. However, our models suggest an independent effect even after adjustments for diabetes medication adherence: number of daily insulin injections, appointment “no show” rate, performance of annual ophthalmology exams, and markers of severity (a hospitalization or emergency room visit during the baseline year). In fact, adjustment for these potential confounders resulted in only minimal changes in the point estimates for the effect of SMBG, suggesting a robust relationship. Finally, the study population is large, ethnically diverse, and socioeconomically and demographically representative of the surrounding geographical region (19
These observational findings are consistent with a short-term benefit of initiating SMBG practice for all patients but a maintained benefit only for pharmacologically treated patients. As with all observational research, bias poses a potential threat to the validity of these findings. However, concerns should be lessened given that the effect size and direction observed in this study closely match those reported by the two most recent and comprehensive meta-analyses (13
) of randomized controlled trials of SMBG. This large observational study complements extant experimental studies. Additionally, it examines something not previously examined: the impact of SMBG in new users versus ongoing users.
Current SMBG guidelines generally encompass only ongoing practice and recommend minimal (if any) monitoring for patients not treated pharmacologically and thus may have missed an important teaching opportunity by failing to make special recommendations for those who initiate SMBG. The benefit of SMBG may be further increased by better integrating SMBG practice into an overall program of health education (promoting patient-level behavioral modifications in response to SMBG readings) and therapeutic decision making (22
). Research is needed to develop specific clinical guidelines for patients regarding optimal timing and frequency of SMBG initiation and practice maintenance and to help clinicians better integrate patient's SMBG records into the therapeutic decision-making process.
Evidence-based practice recommendations are rarely based on observational study findings, relying instead almost exclusively on randomized trials. Unfortunately, authors of the existing meta-analytic studies reported that few of the reviewed randomized studies were of high quality (i.e., design flaws, under-powered, or inadequate follow-up). The controversy surrounding this costly practice will not likely be resolved until a large-scale, well-designed randomized trial of SMBG adequately informs us about its effectiveness and cost-effectiveness. In the meantime, the similar finding of effect in the meta-analyses of existing randomized trials and this observational study is compelling enough evidence to warrant support of SMBG for motivated patients who are appropriately educated in its use.