The discovery of effective tuberculosis (TB) drugs, starting with streptomycin in 1944, resulted in dramatic declines in deaths due to the disease. In the First Nations population of Saskatchewan, the TB death rate fell from 63.1 per 10,000 people in 1933 to 21.0 per 10,000 in 1953  (Fig. 7). Untreated TB often has a protracted course, with estimates of the average time to death or self-cure in the range of 2 to 3 years . Our a priori expectation was that the length of sanatorium admission would decrease after the discovery of effective TB drugs. Instead, we found evidence suggesting that inpatient treatment of First Nations TB patients expanded after novel drug treatments became available.
First Nations TB patients in the mid twentieth century endured prolonged hospitalizations: median 396 (range: 3–3017) days for admissions occurring between 1933 and 1959. Over the course of the study interval the number of first admissions increased, and the length of first admissions generally increased and then stabilized in the late 1950s. We also found that readmission rates increased during the data collection period, from approximately 35% of all admissions in the mono- and dual-therapy eras to over 50% in the triple-therapy era. Admission length fluctuated across eras associated with progressively more effective TB treatments: a demographically similar patient would expect to be admitted for 32 (10.2%) fewer days in the mono-therapy (streptomycin) era compared to the pre-chemotherapy era, suggesting that use of the drug was associated with a shortened interval to clinical cure and/or death. Reversing that trend, patients in the dual-therapy (streptomycin and PAS) and triple-therapy (streptomycin, PAS, and INH) era were admitted on average 76 (24%) and 61 (19%) days longer than in the pre-chemotherapy era respectively. This could represent evolving goals of treatment and/or changes in the standard for what constituted clinical cure. The lower death rate and rising readmission rate in the post-chemotherapy era indicate that advances in TB treatment helped keep patients alive, even if, as is the case with monotherapy , the treatment may not have been effective enough to cure the patient. We note that some patients admitted in the mono-therapy era, which as defined is only 2 years, likely changed regimens during their admission given that median admission lengths during that era approached 1 year (Fig. 2). However, considering all eras prior to effective therapy, we find significant change in admission before and after the advent of triple-therapy: after 1952, and the discovery of INH, we find a significant increase in the proportion of patients who completed therapy at these three sanatoria, indicative of the effectiveness of triple-therapy (Table 3).
We also found that admission length varied as a function of smear status on admission, institution of admission, and in response to the interaction between age and sex. This variation is expected given what we know of TB clinically (longer treatment is required for smear positive pulmonary disease ) and from the underlying data (Fig. 3, Table 1). The above predictor variables collectively accounted for as much variability in admission length as year of admission. The most striking difference in admission length is across smear status, with patients smear positive on admission staying on average 57% longer than demographically similar smear negative patients. Smear positivity is an indicator of disease severity  and this result indicates that our model performs as expected. Though not statistically significant in the model, we find that treatment length varied across diagnoses, with the longest predicted treatment length (520 days) associated with the most severe diagnosis, disseminated disease.
The post-World War II period in Western Canada was a period of intensification of TB prevention and care efforts that included active surveillance of the entire population, suggesting that delays between disease onset and sanatorium admission may have been relatively brief during this interval  and total disease duration may not have been markedly longer than the treatment duration estimated here. Frequently cited estimates obtained for modern, post-chemotherapy data, of the duration of untreated, active TB infection before self-cure or death are approximately 2 years . However, a meta-analysis of pre-chemotherapy studies of tuberculosis incidence, prevalence, and mortality found that the average duration of untreated TB in HIV-negative patients was approximately 3 years until death or self-cure . From the unmodeled data, we find median admission length was 297 days (9.8 months) in the pre-chemotherapy era, 212 days (7.0 months) in the mono-therapy era, 518 days (17.0 months) in the dual-therapy era, and 437 days (14.4 months) in the triple-therapy era. Model-adjusted median admission lengths from the regression analysis are: 316 days (10.4 months) in the pre-chemotherapy era, 284 days (9.3 months) in the mono-therapy era, 392 days (12.9 months) in the dual-therapy era, and 377 days (12.4 months) in the triple-therapy era. This suggests that in the nascent antibiotic era, the effect of TB treatment on shortening disease duration was modest. It is striking that greater than 1 year of hospitalization was deemed necessary to achieve clinical cure, even in the triple therapy era. By comparison, current clinical practice guidelines recommend 26 weeks (6.5 months) of combination antibiotic therapy for the treatment of drug susceptible pulmonary TB .
The middle decades of the twentieth century were a time of transition in TB treatment. In 1948, the World Health Organization (WHO) began a domiciliary drug program that recommended TB treatment take place in an outpatient setting . However, fully outpatient treatment was not achieved until the next decade: treatment guidelines from the 1950s, including recommendations from a clinical study of TB treatment in 1953 , recommended triple-therapy antibiotic courses of 18–24 months to achieve cured status . It was recommended that treatment be administered in an inpatient setting at least until the patient was no longer infectious . By 1959, prominent TB researchers were confident TB could be treated from start to end on an outpatient basis  and by 1965, TB treatment in Canada transitioned from inpatient treatment to a domiciliary drug program . We find that the First Nations patients were treated as inpatients during this transition decade for what appears to be the entire recommended course of treatment (average 16.3 months). Sanatorium treatment declined rapidly in Canada starting in the mid-1950s until most sanatoria were closed in the late 1960s [1, 2].
We found that admission lengths increased as treatment methods became more effective [11, 13, 14] in concert with an expansion in the total number of First Nations patients admitted to sanatoria. Medical historians have pointed to striking differences in TB control policy applied to First Nations populations in this era [1, 2, 4]. One possible explanation of our observation of intensified inpatient treatment of First Nations patients with TB at the advent of the chemotherapy era is a shift to outpatient TB treatment for the non-First Nations population , made possible by the discovery of PAS in 1946 and INH in 1952 , which were taken orally, as opposed to streptomycin, which was taken intravenously . As non-First Nations patients were shifted to outpatient treatment and the sanatorium system risked fading to obsolescence, a proposal to open 1390 newly vacated beds to First Nations patients was put forth in 1945 . Prior to the 1940s, sanatorium treatment was not widely available to First Nations TB patients and was not pursued vigorously until the 1945 proposal, yet beds did not become available until 1950 . Despite effective chemotherapy and the growing feasibility of outpatient treatment, institutionalization and isolation continued to be primary goals of First Nations healthcare policy in Saskatchewan . Medical historians have hypothesized that such policies were devised to protect the interests of health care institutions and their employees as improvements in TB prevention and care made it difficult to justify ongoing allocation of resources to specialized hospitals and personnel . Rising readmission rates (Fig. 6) confirm that new beds were indeed becoming available during the dual- and triple-therapy eras and also indicate that the antibiotic regimens were increasingly effective, with more patients surviving long enough to be readmitted. Consequently, the peak sanatoria admissions year for the entire Saskatchewan First Nations population occurred in 1953 . Our analyses reveal a similar pattern, with both first and overall admissions peaking in 1954 at the three sanatoria in this dataset (Fig. 5). This coincides with increased availability of triple treatment and sanatorium beds made available to the First Nations population.
It is not surprising that we find overall admissions low and death rates high before the mid-1940s. It was known at the time that pre-chemotherapy therapies and monotherapy were not very effective [9, 16,17,18]. Dual therapy was slightly more effective, though the death rate was still higher than for triple therapy [13, 14, 19]. By the mid-1950s it was well established that triple therapy was effective, with INH being the most effective of the three antibiotics [13, 14, 20]. Medical historians have hypothesized that isolation in sanatoria was a key driver in decreasing TB incidence in this population . We find some support for this hypothesis as death rates started decreasing in the mid-1940s, before more effective triple-therapy was available (Fig. 7). Medical histories [15, 21] and firsthand accounts describe the severe hardships associated with TB treatment for First Nations during this era [22, 23]. The difficulty of enduring such prolonged hospitalization is highlighted by our finding that as treatment length became regularized and admission lengths increased during the 1950s, the proportion of patients lost to follow up, categorized in this study as those who left against medical advice or were evicted, increased (Fig. 5). Additionally, it appears that most TB deaths occurred in the hospital, rather than at home, for patients originally hospitalized (Table 3, Fig. 5), highlighting the arduous and isolating nature of TB treatment in this era.
Our analyses reflect length of treatment, rather than length of illness, so our results reflect the effect of changing treatment methods on treatment length, rather than illness duration. Because the sanatorium records do not contain notes of specific treatment received, we have classified each treatment era by the year the drug was discovered. This may have induced classification error in individuals admitted near the cut-off years. We believe any classification error was small, since the average patient who completed therapy was admitted more than a year, well into the new treatment era and likely received at least some of the new antibiotic. The pooled regression model produced poor fit statistics, however it is unclear if these are an artifact of the pooling method or reflective of the model itself. We expect some poor fit due to nonlinearity of the data, and the associations from the model are still clear.